feat(desktop): read replies aloud (auto-TTS) composer toggle

[OutThisLife]

Summary

Adds a one-click "Read replies aloud" toggle to the desktop composer — pure TTS for people who want spoken responses but don't want dictation or the full hands-free voice conversation. Type as normal; when it's on, every completed assistant reply is read aloud.

The toggle drives the existing voice.auto_tts config key (already in Settings → Voice and honored by the messaging gateway), which the desktop chat renderer previously never consumed. So the composer button and the Settings switch are now one source of truth, and the preference persists across restarts.

UX

• New speaker icon in the composer controls, between the mic and the send/voice button.
  • Off → muted-speaker (VolumeX), ghost. On → speaker-waves (Volume2), accent-filled, aria-pressed. Mirrors the existing muted-mic pressed state.
• Complements dictation + conversation rather than replacing them (a TTS-only user simply leaves the mic alone).
• Reuses the existing "Speaking… [Stop]" playback pill for feedback.

Behavior

• Does not read the reply already on screen when toggled on, and won't re-read when opening an old chat — only replies that arrive afterward.
• Never overlaps clips: a reply that lands mid-playback is held and spoken the instant the current one ends (re-checks on both the message-complete and playback-idle edges). Always reads the latest reply, so a backlog collapses to the newest.
• Stays out of the way while a full voice conversation is active (that mode speaks replies itself) and shares lastSpokenId dedupe so nothing is spoken twice.

Test plan

• Toggle on, send a message → reply is read aloud.
• Send a follow-up while a reply is still playing → newer reply is read once the current clip finishes.
• Toggle reflects / persists via Settings → Voice "Read Responses Aloud" and across restart.
• Opening an existing chat with auto-speak on does not read its last reply.
• Full voice conversation still behaves normally (no double-speak).

[alt-glitch]

This was generated by AI during triage.

Duplicate of #43845 — both add a desktop composer 'Read replies aloud' toggle that wires the existing voice.auto_tts config key into the desktop chat renderer to auto-read completed assistant replies. #43845 (filed earlier, implements spec issue #43767) is the canonical PR for this feature; this is the same mechanism on the same surface. Flagging for the maintainer to pick one.