Markup v3: native accept/reject on suggestions (post-multiplayer evaluation)

[gbasin]

Per-suggestion accept/reject buttons on markup cards/docs. Parked: Apply-with-agent (#261) covers resolution end-to-end; the CriticMarkup→ProseMirror round-trip parser (in flight) removes one blocker. Evaluate only if collaborators ask to resolve markup without an agent. Requires: suggestion identity, applying edits to originals (message.edited / artifact version), per-span UI.

[gbasin]

Design locked 2026-07-03 (with Gary), still parked pending real multiplayer demand:

• Blocker cleared: the CriticMarkup→ProseMirror round-trip parser shipped in feat(surface): markup round-trip parser, latest-text-wins refs, mobile webview authoring + files chips #274 — marked-up docs reopen as live suggestions, so per-suggestion resolution is now purely a product-surface build.
• Surface: editor review mode only. Open the marked-up doc → per-suggestion accept/reject affordances; resolving commits the next artifact version. No card-inline buttons (avoids span-identity plumbing on non-editor surfaces).
• Write-back: artifact only. Accepting produces a clean artifact version; the original chat message is never edited. Anyone can copy the resolved text back into a message manually.
• Permissions: anyone in the channel, matching the uniform-editing model already shipped (anyone can retract anyone's suggestion in the editor).
• Mechanics (markup-as-content makes this cheap): accept = apply the span's edit and strip tokens; reject = strip tokens keeping the original; each resolution is just the next version. No suggestion-identity store needed in the editor — spans are live marks after parsing.

Trigger to build: collaborators asking to resolve markup without an agent.