fix: stabilize dashboard navigation recovery

[numnx]

What changed

• Prevent chat draft hydration from pairing a newly selected project with a thread retained from the previous project.
• Rebase the shared realtime cursor to the server's authoritative acknowledgement after restart.
• Clear unrecoverable realtime cursors and serialize navigation-driven subscription changes until acknowledgement.
• Update canonical and published documentation for draft ownership, recovery operations, and the current WebSocket protocol.

Root cause and impact

Project navigation could briefly combine the new project id with the previous thread id, causing draft GET/PUT ownership failures. Separately, the realtime client max-merged sequence acknowledgements, so a lower post-restart server watermark could never replace the stale browser cursor. Every navigation scope change then triggered another snapshot recovery and a fan-out of REST refreshes, which could make route tabs appear frozen.

Validation

• pnpm run ci
  • dependency audit passed
  • typecheck/lint passed
  • backend coverage: 5,262 passed, 7 skipped
  • dashboard: 2,100 passed, 2 skipped
  • production build passed
• Targeted dashboard regressions: 31 passed
• Playwright route smoke and rapid navigation stress: 2 passed
• Fixed build on port 4444: rapid in-app route cycle completed with no page errors or HTTP failures
• Published docs sync check passed

Risk and rollback

Risk is limited to dashboard draft-context selection and realtime subscription handshakes. Roll back commit d5284c048 if scope acknowledgements fail to progress or draft restoration regresses. REST resource snapshots remain the recovery source throughout.