Official-client parity: file-id routing, failure backoff, push-aware intervals, heartbeat (#183 #184 #185 #186)

[gnacho]

Benchmarked against the official Nextcloud desktop client (nextcloud/desktop) and ported four robustness/performance improvements. Closes #183, #184, #185, #186.

#183 — Resolve notify_file_id and sync only the affected folder

The push client treated every notify_file_id <id> as a generic change and fanned out RemotePush to every folder of the account. The official client resolves the id against its folder journal (SyncJournalDb::hasFileIds) and schedules only the folder that owns the file.

• push.rs: opt in to listen n_id after authenticated; parse notify_file_id <json-array> into file ids (FileNotification(Vec<i64>)); a legacy notify_file keeps an empty list.
• New core/files_journal.rs: read-only lookup of the external engine's .sync_*.db journal (the same journal nextcloudcmd keeps), answering whether a folder knows any notified id (metadata.fileid compared via CAST(... AS INTEGER)).
• Scheduler::has_file_ids + apply_remote_push: route the hint to only the folders that contain the file; empty ids (legacy) still fan out.

#184 — Per-folder consecutive-failure backoff (10/30/60s)

The official client delays the next run by 10s/30s/60s depending on consecutiveFailingSyncs. Ours re-ran on every trigger.

• consecutive_failing_syncs counter on the scheduler (reset on success/conflict, incremented on failure).
• failure_backoff_seconds() (10/30/60) + schedule_after(): cooldown_finished backs off instead of re-running immediately once 3+ consecutive failures accrue.

#185 — Skip remote-interval polling while push is active

slotEtagPollTimerTimeout in the official client only polls accounts whose push is not ready. Ours armed RemoteInterval regardless.

• remote_push_ready flag on the scheduler (set_remote_push_ready), wired from store_push_state: when the account's push is Connected, a RemoteInterval trigger is dropped.

#186 — WebSocket heartbeat + capped auth retries

The official client sends a 30s ping, detects a missing pong within its timeout, and caps auth attempts at 3.

• push.rs: track last_activity in the worker loop; a connection with no inbound text/ping/pong for PUSH_HEARTBEAT_TIMEOUT_MS (75s) is torn down and reconnects (no silent zombie).
• Cap consecutive failed auth attempts at MAX_PUSH_AUTH_ATTEMPTS (3); once exhausted the channel parks in AuthRequired instead of reconnecting in a loop. The streak resets on Authenticated and on coming back online.

Testing

685 tests (added 8), clippy clean, fmt applied.

New tests cover notify_file_id parsing, journal file-id membership, the failure-backoff table, and dropping RemoteInterval while push is ready.