room_view: mirror face_recognized into per-device state

[BrettKinny]

Summary

Follow-up to PR #102. The room_view branch in `routes/vision.py` calls `state.broadcast()` on roster match but skipped the `state.update_state()` call that `POST /api/perception/event` normally pairs with broadcast. Result: `last_face_id` and `last_face_recognized_t` never landed on per-device state, even though the bus event fired correctly.

Why

Caught during the #44 LED bench sweep immediately after #102 merged. Brett walked in, named-greet fired correctly (event reached FaceGreeter), but the right-ring pixel 6 wouldn't hold green past the firmware's 4s timeout because `face_identified_refresher` reads `last_face_id` to decide whether to re-fire `set_face_identified`.

Live state confirmation pre-fix (snapshot from `curl /api/perception/state`):
```
"last_name_greet_t": {"brett": 1779532034} ← FaceGreeter saw the event
…
← no last_face_id field
```

Fix

Mirror what `POST /api/perception/event` does:
```python
state.update_state(device_id, "face_recognized", data, ts) # ← added
state.broadcast(PerceptionEvent(...))
```

Other affected readers that were silently no-op'ing:

• `perception/snapshot.py` (`last_face_id` for talk-turn PerceptionSnapshot "Speaking with: …" hint to the LLM)
• `state.get_fresh_face_id()` (canonical fresh-identity accessor)

Test plan

• `pytest tests/test_routes_vision.py -q` — 10 passed
• `pytest tests/ -q` — 198 passed (no regressions)
• Existing happy-path integration test extended with assertions on `state.state[device]["last_face_id"] == "brett"` and `last_face_recognized_t > 0` — locks in the invariant so the regression can't reappear silently
• Live walk-in: pixel 6 should now hold green while face_present=True (refresher fires every 3s while identity is fresh)

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes a missing per-device perception state update in the room_view roster-match path of POST /api/vision/explain, ensuring last_face_id and last_face_recognized_t are written when a face_recognized bus event is emitted (matching the behavior of POST /api/perception/event). This unblocks downstream readers like face_identified_refresher, snapshots, and get_fresh_face_id() that depend on per-device state rather than just the broadcast event.

Changes:

• Update routes/vision.py to call state.update_state(..., "face_recognized", ...) before broadcasting the face_recognized event.
• Extend the existing room_view happy-path integration test to assert last_face_id and last_face_recognized_t are set.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
dotty-behaviour/routes/vision.py	Mirrors /api/perception/event by updating per-device state before broadcasting face_recognized for room_view roster matches.
dotty-behaviour/tests/test_routes_vision.py	Adds assertions to lock in the invariant that roster matches populate last_face_id / last_face_recognized_t.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.