Reconnect and missed-event recovery

## Summary
Implement reconnect and missed-event recovery so realtime behavior remains trustworthy through disconnects and refreshes.

## Scope
- Recover from websocket disconnects and browser refreshes without losing obvious user context.
- Reconcile missed messages, unread state, and channel/thread state after reconnect.
- Handle duplicate or out-of-order event delivery safely.
- Provide visible recovery behavior instead of silent stale UI.

## Acceptance Criteria
- [ ] Refreshing or reconnecting restores the user to a correct guild state without missing recent messages.
- [ ] Unread counts, mentions, and last-read behavior remain correct across reconnects.
- [ ] Duplicate or replayed events do not corrupt transcript state.
- [ ] The product communicates reconnecting/recovered states when the interruption is user-visible.

## Operational Notes
- This issue is about user-visible reliability, not only transport correctness.
- Recovery paths should keep working under higher load and future multi-node delivery.

No files, tests, or entry points are identified in the issue. Start by mapping the websocket reconnect, missed-message, unread-state, and channel/thread recovery paths. Done means refreshes and reconnects restore correct state, duplicate or out-of-order events are safe, and user-visible recovery states are communicated.

Written by the indexing model from the issue text.