…rror recovery, and TTS discovery

- Add voice mode state machine: Recording → Processing → Waiting → Generating → Playing → Recording loop
- Add audio error recovery with blob retention, retry, and discard
- Add audio cues (mic-on.wav, mic-off.wav) for recording state transitions
- Add TTS discovery prompt for first-time users on supported platforms
- Extend RecordingOverlay with 6 voice mode states (recording, processing, error, waiting, generating, playing)
- Update TTSContext with isGenerating, speakAndWait, cancelGeneration, and sequence ID tracking
- Update TTSButton with generating animation state
- Add voice mode exit on chat switch, new chat, and manual stop
- Highlight mic button when voice mode is active

Closes #458

Co-Authored-By: marks <markskram@protonmail.com>

Co-Authored-By: marks <markskram@protonmail.com>

…VoiceDiscard

- Add recordingStartTimeRef to track when recording starts (RecordRTC has no startTime property)
- Add startRecordingRef to avoid stale closure in handleVoiceDiscard's empty dependency array
- capturedDuration now correctly reflects actual recording time instead of always being 0

Co-Authored-By: marks <markskram@protonmail.com>

… for savedDuration

- Replace all 3 startRecording() calls in voice continuation effect with startRecordingRef.current()
- Change recordingDuration || undefined to recordingDuration ?? undefined (both inputs)
  to correctly show 0-second durations in error UI

Co-Authored-By: marks <markskram@protonmail.com>

…g on empty blob

- isTTSPlatform now checks isTauri() && isIOS() to match TTSContext behavior
- Empty blob in voice mode now calls startRecordingRef.current() to restart

Co-Authored-By: marks <markskram@protonmail.com>

…ssage/handleTTSDiscovery

- exitVoiceMode: check recorderRef.current instead of isRecording state to avoid stale closure in event handler effect
- Add handleSendMessageRef and handleTTSDiscoveryRef to prevent stale closures in transcribeAndSend
- transcribeAndSend now calls through refs for latest handleSendMessage and handleTTSDiscovery

Co-Authored-By: marks <markskram@protonmail.com>

…e to avoid stale closure

These functions are idempotent, so guarding on ttsIsGenerating/ttsIsPlaying
was unnecessary and caused stale closures when exitVoiceMode was captured
in the event listener effect.

Co-Authored-By: marks <markskram@protonmail.com>

…on generation error

- Set voiceState(null) after successful transcription in non-voice mode to dismiss overlay
- Add errorRef to track error state without stale closures
- Voice continuation effect checks errorRef.current and exits voice mode on error
  instead of speaking stale assistant message

Co-Authored-By: marks <markskram@protonmail.com>

…de exits

Without re-throwing, speakAndWait always resolves normally even on TTS failure,
causing the voice mode continuation effect's .catch() to never fire and creating
an infinite loop: TTS fails → recording restarts → repeat.

Co-Authored-By: marks <markskram@protonmail.com>

…n TTSButton

speak() is used by TTSButton which doesn't have try/catch. Only speakAndWait
(used by voice mode loop) needs error propagation to exit on TTS failure.

Co-Authored-By: marks <markskram@protonmail.com>

Devin Review round 12:
- TTSContext: store audioContextRef immediately after AudioContext creation
  so stopPlayback() can close it if decodeAudioData or other operations
  between creation and the old ref assignment throw.
- TTSContext: throw Error('no_speakable_text') instead of silently returning
  when preprocessTextForTTS strips all content (e.g. code-only responses).
- UnifiedChat: catch 'no_speakable_text' in both voice continuation effects
  and restart recording instead of exiting voice mode, so the user gets
  audio feedback (mic-on cue) rather than a silent mic activation.

Co-Authored-By: marks <markskram@protonmail.com>

…extRef cleanup

Devin Review round 13:
- UnifiedChat: in startRecording's catch block, call exitVoiceMode() when
  voice mode is active so the UI doesn't show 'Recording' overlay when no
  recording is actually happening (mic permission denied, device busy, etc.)
- TTSContext: only null audioContextRef.current in the staleness check if
  it still points to this call's AudioContext, preventing a concurrent
  speakInternal call's ref from being clobbered.

Co-Authored-By: marks <markskram@protonmail.com>

…, compact layout gap, TTS auto-play after download

Bug 1: Audio cues not playing on iOS - AudioContext starts in 'suspended' state
on iOS; added ctx.resume() before fetching and playing the wav file.

Bug 2: Previous recording duration flashes on re-entry - setDuration(0) in
useEffect fires after the first render with stale state. Added synchronous
state reset during render when effectiveState transitions to 'recording'.

Bug 3: Compact playback overlay layout imbalanced - waveform too near top,
status text too near bottom. Reduced gap from gap-6 to gap-2 in compact mode.

Bug 4: TTS auto-play after model download not triggering - when ttsStatus
transitions to 'ready' mid-voice-loop, the user may already be recording.
Now stops active recording before transitioning to 'generating' state.

Co-Authored-By: marks <markskram@protonmail.com>

…c decoding window

Co-Authored-By: marks <markskram@protonmail.com>

The 'playback' session type is incompatible with getUserMedia on iOS.
Now saves the previous session type before setting 'playback' and
restores it in onended/catch/error paths so mic activation works.

Co-Authored-By: marks <markskram@protonmail.com>

…e debug buttons

Co-Authored-By: marks <markskram@protonmail.com>

Co-Authored-By: marks <markskram@protonmail.com>

…oice loop

Co-Authored-By: marks <markskram@protonmail.com>