fix(desktop): re-arm voice chat after replies

[marcdordal]

Summary

• capture the playback sequence after voice playback starts, so Hermes does not confuse its own stale-audio reset with a user Stop action
• re-arm Start Voice after a normal streaming reply
• cover the playback-sequence regression

Closes #73649.

Test Plan

• npm test -- --run src/app/chat/composer/hooks/use-voice-conversation.test.tsx
• npm run typecheck
• npm run lint -- --quiet src/app/chat/composer/hooks/use-voice-conversation.ts src/app/chat/composer/hooks/use-voice-conversation.test.tsx
• Manual Desktop test: Start Voice accepted a second request after the first reply finished

[alt-glitch]

This was generated by AI during triage.

Duplicate of #73880: both move the voice-playback sequence baseline past the hook's internal playback reset/start so a normal reply re-arms the microphone rather than looking like an explicit Stop.

[teknium1]

Thanks for the focused renderer regression fix. The current-main premise is valid: openLiveSpeech() records its baseline before startSpeechStream() at apps/desktop/src/app/chat/composer/hooks/use-voice-conversation.ts:487, while startSpeechStream() later calls stopVoicePlayback() after asynchronous URL resolution at apps/desktop/src/lib/voice-playback.ts:324-332.

Problems

• The new post-await startSpeechStream() baseline at apps/desktop/src/app/chat/composer/hooks/use-voice-conversation.ts:509 absorbs an explicit Stop made while stream discovery is pending. The helper's normal reset and that user Stop are both included in the recorded baseline, so the later sequence comparison cannot suppress re-arm.
• The new test covers normal streaming completion only. The fallback path changed too, and explicit-Stop behavior remains untested.

Suggested changes

• Preserve the entry sequence and allow only the one expected setup increment after startSpeechStream() resolves; stop and settle without re-arming if an additional increment occurred.
• Cover Stop during setup, Stop during active playback, normal fallback re-arm, and Stop before fallback begins.

Automated hermes-sweeper review.

[teknium1]

startSpeechStream() awaits URL discovery before it performs its one expected stopVoicePlayback() increment. A user Stop during that await is absorbed by this post-await baseline, so later settling cannot distinguish it from normal setup and re-arms the mic. Preserve the entry sequence and treat more than the expected one increment as an external Stop.