feat(reticulum): Ratspeak-parity LXMF voice memos

docs/agents/chat.md

@@ -4,7 +4,7 @@ Deep subsystem reference for AI assistants. Open this when a task touches the Ch
 
 - **Components:** `ChatPanel.tsx` (channel/DM UI) + shared `ChatComposer.tsx` (drafts, mentions, chunking, spellcheck, emoji; also used by `RoomsPanel.tsx`). Reticulum DM **Share as paper** / **Scan paper** via `ChatDmPaperControls.tsx` + `createReticulumPaperMessage.ts`. Scroll-at-bottom helper: `chatScrollUtils.ts` (`getDistFromChatBottom`).
 - **Composer limits / send cadence:** `chatComposerLimits.ts` — `getMaxChunks(protocol)` via `ProtocolCapabilities.composerMaxChunks` (MeshCore = 1: no outbound `[i/N]` split; `splitChatMessage` returns `null` when text needs more than one packet), room payload via `getMeshcoreRoomPayloadLimit`, `computeComposerLimitStatus` phases (`warn` surfaces a single-packet ⓘ hint; `overMaxSingle` disables send and shows a `role="note"` callout). MeshCore also gets a **non-blocking** ~5s "sending too fast" advisory (`role="status"`, dismissible) from an app-wide clock in `meshcoreSendRateNotice.ts` (`recordMeshcoreSend` / `isMeshcoreSendTooFast`, `MESHCORE_FAST_SEND_WARN_INTERVAL_MS`). Every **live** MeshCore send advances the clock and can show the advisory — text (`handleSend`), GIF, and share-location in `ChatComposer`, plus outbox drain (`useChatOutbox.ts`). Legacy MeshCore outbox rows with `groupTotal > 1` or `[i/N]` payloads are quarantined (`blocked`) on drain instead of being transmitted. Inbound multi-part `[i/N]` merge is unchanged. i18n: `chatPanel.composeLimit.meshcoreSingleNotice.*`, `chatPanel.meshcoreFastSend.warning`, `chatPanel.outboxLegacyMultipartBlocked`. See [`meshcore-meshtastic-parity.md`](../meshcore-meshtastic-parity.md).
-- **Payload / links:** `ChatPayloadText.tsx` — mention highlighting, search marks, URL linkification; link previews via `chat:fetchLinkPreview` (`src/main/fetchLinkPreview.ts`): Open Graph for HTML pages; **YouTube** watch/shorts/youtu.be via oEmbed + thumbnail; **direct image URLs** (path extension via `chatDirectImageUrl.ts` or raster `Content-Type`) return `kind: 'image'` and render as inline embeds (`ChatInlineImage` / `DirectImageEmbed`); OG/YouTube use card layout. Security: DNS-pinned undici `Agent`, private/loopback blocked, magic-byte MIME sniff (`safeRasterImageMime.ts`), HTTPS-only image embeds, 10s fetch / 3s DNS, 64 KiB HTML cap, **2 MiB** image fetch cap (256 KiB cache payload cap), LRU caches, single-flight dedup (renderer map capped). Previews load even when scrolled up. LXMF attachment rasters: `chat:readReticulumAttachmentAsDataUrl` (`reticulum-attachment-image.ts`; path jail, magic-byte MIME, SVG rejected, 2 MiB, IPC rate limit) → `ReticulumAttachmentLine`. Reply quotes: `replyPreview.ts`.
+- **Payload / links:** `ChatPayloadText.tsx` — mention highlighting, search marks, URL linkification; link previews via `chat:fetchLinkPreview` (`src/main/fetchLinkPreview.ts`): Open Graph for HTML pages; **YouTube** watch/shorts/youtu.be via oEmbed + thumbnail; **direct image URLs** (path extension via `chatDirectImageUrl.ts` or raster `Content-Type`) return `kind: 'image'` and render as inline embeds (`ChatInlineImage` / `DirectImageEmbed`); OG/YouTube use card layout. Security: DNS-pinned undici `Agent`, private/loopback blocked, magic-byte MIME sniff (`safeRasterImageMime.ts`), HTTPS-only image embeds, 10s fetch / 3s DNS, 64 KiB HTML cap, **2 MiB** image fetch cap (256 KiB cache payload cap), LRU caches, single-flight dedup (renderer map capped). Previews load even when scrolled up. LXMF attachment rasters: `chat:readReticulumAttachmentAsDataUrl` (`reticulum-attachment-image.ts`; path jail, magic-byte MIME, SVG rejected, 2 MiB, IPC rate limit) → `ReticulumAttachmentLine`. LXMF voice memos (`hasReticulumVoiceMemo`): Chat DM mic → sidecar `/api/v1/voice/memo/*` via `electronAPI.reticulum.voiceMemo.*`; playback via `chat:readReticulumAttachmentBytes` (`reticulum-attachment-audio.ts`; OggS sniff, 256 KiB) → `ReticulumVoiceMemoLine`. Reply quotes: `replyPreview.ts`.
 - **Storage helpers:** `src/renderer/lib/chatPanelProtocolStorage.ts` — drafts (`mesh-client:drafts:<protocol>`), open DM tabs, last-read, per-view mute (`mesh-client:mutedViews:<protocol>`), starred (`mesh-client:starred:<protocol>`, cap 200), MeshCore flood-scope overrides per chat view (`mesh-client:floodScopeOverrides:<protocol>`, channel or DM `viewKey`).
 - **Notifications:** `src/renderer/lib/chatNotifications.ts` — `playMessageNotification(type)` via Web Audio: `channel` = single 880 Hz pulse (150 ms); `dm` / `reply` = dual pulse (587.33 Hz then 783.99 Hz, 50 ms each, 35 ms gap). Resumes suspended `AudioContext` when the window is hidden/minimized. Type selection in `chatUnreadCounts.ts` (`resolveChatNotificationType`, `pickAudibleNotificationType`; batch priority reply > dm > channel). **ChatPanel** plays when the user is on Chat but reading another view; **App** plays for other panels / backgrounded window (avoids double beep). Meshtastic hidden-window desktop notifications are visual-only (`silent: true` in `meshtasticRouterSideEffects.ts`); typed Web Audio from App owns sound. Global mute `mesh-client:notifMuted`; per-view mute in `mutedViews`. Main-process **tray** icon shows unread when chat or MeshCore Rooms traffic arrives while backgrounded (`src/main/index.ts` `buildTrayIcon`).
 - **Meshtastic dedup:** `meshtasticMessageDedup.ts` — merges delayed RF/MQTT duplicates (**10-minute** content window) in `useMeshtasticRuntime` ingest.

docs/agents/reticulum.md

@@ -28,8 +28,9 @@ Deep subsystem reference for AI assistants. Open this when a task touches the Re
 - **Nomad errors:** `lib/nomad/nomadPageErrorHumanize.ts` (sidecar error codes → i18n); LinkClient Nomad overlay in `reticulum-sidecar/patches/`
 - **Nomad request bodies:** MessagePack `field_*`/`var_*` encode/decode lives in rsNomad `nomad-core` (`encode_request_fields` / `decode_request_fields`); sidecar `nomad_request_payload.rs` only unwraps HTTP base64 JSON before calling encode
 - **LXST voice:** `hasLxstVoice` gates Call buttons (Peers + Chat DM). Session helpers in `reticulumVoiceSession.ts` (dial/answer/hangup + mic PCM); UI store `reticulumVoiceStore.ts`; overlay `ReticulumVoiceOverlay` (App mount). Dedicated IPC `reticulum:voiceSendAudio` + push channel `reticulum:voiceAudio` (`/ws/voice`; preload `onVoiceAudio`); control via `electronAPI.reticulum.voice.*`. Runtime WS: `voice.update` / `voice.incoming` / `voice.stats` / `voice.terminated` / `voice.error` (errors should carry `link_id` when known; match by link/generation/remote). **Establish-only media:** Answer warms AudioContext; mic capture/TX starts only after `established`; sidecar soft-drops pre-establish PCM (`not_established`). Outbound progress tones: dial → peer DTMF fold → UK double-ring (`reticulumVoiceCallTones.ts` / `reticulumVoiceOutcome.ts` / `reticulumVoiceFeedback.ts`); media-start coalesces by `callGeneration` to avoid Answer mic thrash. Terminal reasons: treat sidecar `established`/`terminated` as completed (not fail).
+- **LXMF voice memos:** `hasReticulumVoiceMemo` gates Chat DM mic (not LXST Call). Sidecar encodes Opus/`AM_OPUS_OGG` via `/api/v1/voice/memo/{start,audio,stop,cancel}` + dedicated IPC `electronAPI.reticulum.voiceMemo.*` (`reticulum:voiceMemoStart` / `voiceMemoSendAudio` / `voiceMemoStop` / `voiceMemoCancel`; proxy blocked). Capture resamples to 24 kHz / 60 ms frames and serializes PCM IPC before stop. Ogg capped ~240 KiB under the 256 KiB LXMF field / default PN deposit. Key files: `reticulumVoiceMemo.ts`, `sendReticulumVoiceMemo.ts`, `reticulumVoiceMemoStore.ts`, `reticulum-attachment-audio.ts`, `ReticulumVoiceMemoLine.tsx`. Ingest stamps `FIELD_AUDIO` → attachment jail + playback (`chat:readReticulumAttachmentBytes`). Oversize-for-PN emits `message_too_large_for_propagation` (notice toast + Direct-only badge — never a PN-outage toast). Tap mic to record, tap again to send; Esc/DM-switch cancels; mic stays visible while recording even if the draft is non-empty.
 - **LRGP games:** `hasLrgpGames` gates Games tab + Challenge (Peers / Chat DM). Sidecar `games_session` + `LrgpStore`; companion `games_outbound.db` persists last envelope + `delivery_state` (LXMF outbound bridge → session chips / Resend). Dedicated IPC `electronAPI.reticulum.games.*` / `reticulum:games*` (proxy rejects `/api/v1/games/*`); WS `games.update` / `games.action_result`. Parity: [../reticulum-games-parity.md](../reticulum-games-parity.md).
-- **Gating:** `hasReticulumDiscoveryMap` (Map tab); `hasReticulumRemotePanel` / `hasRncpTransfer` (Remote tab + Chat DM rncp); `hasRrcPanel` (RRC tab); `hasLxstVoice` (LXST Call); `hasLrgpGames` (Games); `hasReticulumInterfaceConfig` / `hasReticulumNetworkPanel` / `ProtocolCapabilities`
+- **Gating:** `hasReticulumDiscoveryMap` (Map tab); `hasReticulumRemotePanel` / `hasRncpTransfer` (Remote tab + Chat DM rncp); `hasRrcPanel` (RRC tab); `hasLxstVoice` (LXST Call); `hasReticulumVoiceMemo` (LXMF voice memos); `hasLrgpGames` (Games); `hasReticulumInterfaceConfig` / `hasReticulumNetworkPanel` / `ProtocolCapabilities`
 - **rnsh/rncp:** sidecar `stack/{rnsh_session,rncp_transfer,path_speed,link_task}.rs` + HTTP `/api/v1/rnsh/*`, `/api/v1/rncp/*`, `/api/v1/remote/*`; typed `electronAPI.reticulum.rnsh|rncp|remote`; picker-gated send/fetch paths in `reticulum-remote-paths.ts`; LXMF enable-request sentinel `mesh-client:request-rncp-receive:v1` (`rncpRequestEnable.ts`); peer reply `mesh-client:rncp-receive-dest:v1:<hash>` autofills via `applyRncpReceiveDestShare` (prefer pending from `markRncpReceiveDestSharePending` / `sendRncpRequestEnable`; still apply without pending for older peers); enable-request modal + dest-share side effects deduped by LXMF `message_hash` (`rncpLxmfControlSideEffectDedup`) so catch-up cannot re-fire; already-listening auto-share is once per peer per request-enable cooldown; inbound listener config persists (`rncp_listener_*` in `mesh_client_stack.json`) and restores on live stack start
 - **Runtime:** `useReticulumRuntime`, `lib/sessions/reticulumSession.ts`, `lib/ingest/reticulumIngest.ts`; connect starts sidecar, not `ConnectionDriver` RF — marks **configured** when HTTP + identity ready (live attach may still run); `RETICULUM_CONFIGURED_EVENT` wakes RRC. Cancel/stop is fire-and-forget vs cargo/BLE (`START_ABORTED` checkpoints; next start does not rejoin a doomed promise). LXMF/RRC proxy sends: **15 s** `RETICULUM_IPC_SEND_TIMEOUT_MS`. RRC auto-connect (`useRrcStartupAutoConnect`): ~**500 ms** while hubs pending, ~4 s steady. Sidecar RRC: `rrc_codec` / `rrc_link` / `rrc_session` / `api/rrc.rs`
 - **Diagnostics:** `ReticulumDiagnosticEngine.ts` (Reticulum-native rows; no LoRa hop-goblin semantics) — includes `reticulum/sidecar-unhealthy` (60s grace; HTTP health, not listen-first ready lag), `reticulum/rns-not-ready` / `reticulum/lxmf-not-ready`, `reticulum/propagation-sync-stuck`, `reticulum/propagation-sync-failing` (1h TTL)