feat(desktop): add microphone device selector for voice mode

[EdderTalmor]

Closes #55697

What changed:

• Desktop composer can now select an audio input device before starting voice mode.
• Selected device is persisted across composer/session restarts.

Backend:

• Thread deviceId through the desktop voice recorder hooks and apply it as an exact getUserMedia audio constraint when present.

Frontend:

• Add useMicDevice for device enumeration, selection, and persistence.
• Add a mic device dropdown to the composer controls when multiple inputs are available.

Why:
Users on laptops/docking setups frequently have multiple microphones; without a selector they are forced to use the OS default, which may not be the intended input.

[0xLeathery]

Heads-up: deviceId helps selection; continuous voice still has a VAD SPOF

This composer-level device selector is a clean approach for #76866.

One gap we hit in production use (detailed on #75329): after getUserMedia with the right deviceId, continuous conversation can still drop utterances if the WebAudio level meter dies. Submission is gated on heardSpeech from RMS in use-mic-recorder; a dead AudioContext leaves heardSpeech === false and use-voice-conversation never calls STT — even when the WebM blob is fine.

Since this PR already threads deviceId through use-mic-recorder / use-voice-conversation / use-voice-recorder, it may be the cheapest place to also:

• resume AudioContext / detect unhealthy meter
• fall back VAD from MediaRecorder timeslice chunks
• avoid silently discarding non-trivial recordings

I can provide a small patch atop this branch if useful. Otherwise I’ll leave device selection to this PR / #46916 and keep VAD hardening as a separate follow-up after maintainers pick a primary mic-recorder diff — to avoid a merge pile-up on the same files.

RCA comment: #75329 (comment)

[andrexibiza]

Vox Lockin lane 09 verification receipt (feature PR for #46337):

• Not mergeable as-is: git merge-tree against current main (70db671) conflicts in controls.tsx, composer/index.tsx, and tests/gateway/test_wecom.py (the head also carries an unrelated WeCom commit, 2f049a174f, which should be dropped before merge).
• Desktop-class review: the deviceId plumbing (use-mic-recorder → handle.start({ deviceId }) → getUserMedia exact constraint) is sound and conflicts only because the hook moved on. It overlaps with Add Desktop microphone picker and STT diagnostics #46916 (also a mic-picker feature, also conflicting, currently marked CONFLICTING by GitHub). Both address feature request [Feature]: Add UI for Custom Local STT/TTS and Local Media Generation Providers in Hermes Desktop #46337.
• No bug-class gap: this is an additive feature, not a reliability fix. Recommend the two authors coordinate on one rebased implementation; lane 09 will not duplicate.

Note for the rebase: on current main useMicRecorder.start already guards recorderRef.current and handle.start re-arms with the exact same options object each cycle, so a deviceId captured at conversation start stays stable across re-arms — the feature composes cleanly.