Lifts MediaStream, recording state, transcription buffer, and queue for
non-Assistant captures out of ChatInput's internal VoiceMicButton. The
provider owns the POST /api/transcribe fetch (v1.6 pipeline, unchanged),
exposes idle/listening/speaking state to the top-strip GlobalMicButton,
and queues transcripts captured away from /assistant for PersonalAssistant
to drain on mount.
Per spec sections 4.2 (mic states), 5.5 (voice from non-Assistant modes),
and 10.3 (voice as global affordance). Tests use manual createRoot + act
with a mocked getUserMedia injector to stay deterministic in jsdom.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
14ecbf00bb