fix(participant): guard chunk restart so a dead mic ends as an interruption

[posthog-eu]

Problem

• Participants silently lose the rest of a recording when the mic drops mid-conversation: the recorder stops, but the UI keeps saying it is recording.
• useChunkedAudioRecorder restarts each chunk from the onstop handler, ending in an unguarded recorder.start(). When the mic track dies between chunks (the participant takes a call, or another app grabs the device), start() throws NotSupportedError and the throw escapes onstop.
• Result: isRecording stays true, no further ondataavailable or onstop fires, the suspicious-chunk detection never runs, and onRecordingInterrupted is never called. The participant gets no alert and every second of audio after that point is lost.

Changes

• Wrap the chunk-restart start() in a try/catch that mirrors the existing startRecording failure path: release the mic, drop the recorder, clear isRecording, and raise the same interruption the participant already gets for a dead mic (alert sound + interruption modal via onRecordingInterrupted).
• Only restart calls (from onstop and the restart timer) are handled locally; the initial start() from startRecording still propagates so its own caller resets state and returns false. A new isRestart flag draws that line.
• Guard the previous-recorder stop() on state === "recording" (as stopRecording and pauseRecording do). A restart runs from onstop, where that recorder is already inactive, and stop() on an inactive recorder throws too.
• Factor the alert playback into a small playInterruptionAlert helper shared by both interruption paths.

Testing

• New unit test drives the restart-failure path (mic dies between chunks) and asserts the failure surfaces as an interruption, not a stuck "still recording" state: onRecordingInterrupted fires, isRecording drops to false, and the dead mic is released.
• useChunkedAudioRecorder.test.tsx: 7 passed.

Agent context

• The reported symptom named the start() throw. Reproducing the restart path in the test uncovered a second unguarded throw in the same loop (stop() on the already-inactive previous recorder), fixed the same way.

---

Created with PostHog Desktop from this inbox report.