feat: voice S2S — faster-whisper STT, Kokoro TTS, PTT overlay

Implements full speech-to-speech pipeline (all 4 phases):

Backend (Phase 1):
- services/stt.py: lazy WhisperModel singleton, run_in_executor transcription
- services/tts.py: lazy KPipeline singleton, WAV synthesis at 24kHz/16-bit
- routes/voice.py: /api/voice/status, /voices, /transcribe, /synthesise
- config.py: VOICE_ENABLED, STT_BACKEND, STT_MODEL, TTS_BACKEND env vars
- app.py: load STT/TTS models at startup when VOICE_ENABLED=true
- llm.py: voice_mode + voice_speech_style params inject speak-naturally prefix
- generation_task.py: voice_mode passed through from chat route
- chat.py: "voice" conversation type allowed + excluded from retention cleanup
- pyproject.toml + Dockerfile: faster-whisper, kokoro, soundfile dependencies

Frontend (Phases 2–4):
- composables/useVoiceRecorder.ts: MediaRecorder PTT wrapper
- composables/useVoiceAudio.ts: AudioContext WAV playback wrapper
- BriefingView.vue: Listen button (TTS read-aloud), auto-TTS mode, mic PTT
- VoiceOverlay.vue: global floating PTT button; creates/reuses voice conv;
  full record→transcribe→stream→TTS flow; Space bar hold-to-talk via App.vue
- SettingsView.vue: Voice tab (status badge, speech style, voice/speed)
- App.vue: mounts VoiceOverlay; Space keydown/keyup fires voice:ptt-toggle
- api/client.ts: getVoiceStatus, getVoiceList, transcribeAudio, synthesiseSpeech

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

frontend/src/composables/useVoiceAudio.ts

@@ -0,0 +1,60 @@
+import { ref, readonly } from 'vue'
+
+/**
+ * Audio playback composable wrapping the Web Audio API.
+ *
+ * Usage:
+ *   const { playing, isSupported, play, stop } = useVoiceAudio()
+ *   await play(wavBlob)
+ */
+export function useVoiceAudio() {
+  const playing = ref(false)
+  const isSupported = typeof AudioContext !== 'undefined' || typeof (window as unknown as Record<string, unknown>).webkitAudioContext !== 'undefined'
+
+  let audioCtx: AudioContext | null = null
+  let currentSource: AudioBufferSourceNode | null = null
+
+  function _getCtx(): AudioContext {
+    if (!audioCtx || audioCtx.state === 'closed') {
+      const Ctx = window.AudioContext ?? (window as unknown as Record<string, typeof AudioContext>).webkitAudioContext
+      audioCtx = new Ctx()
+    }
+    return audioCtx
+  }
+
+  async function play(blob: Blob): Promise<void> {
+    stop()
+    const ctx = _getCtx()
+    if (ctx.state === 'suspended') await ctx.resume()
+
+    const arrayBuffer = await blob.arrayBuffer()
+    const audioBuffer = await ctx.decodeAudioData(arrayBuffer)
+
+    const source = ctx.createBufferSource()
+    source.buffer = audioBuffer
+    source.connect(ctx.destination)
+    currentSource = source
+    playing.value = true
+
+    source.onended = () => {
+      playing.value = false
+      currentSource = null
+    }
+    source.start(0)
+  }
+
+  function stop(): void {
+    if (currentSource) {
+      try { currentSource.stop() } catch { /* already stopped */ }
+      currentSource = null
+    }
+    playing.value = false
+  }
+
+  return {
+    playing: readonly(playing),
+    isSupported,
+    play,
+    stop,
+  }
+}

frontend/src/composables/useVoiceRecorder.ts

@@ -0,0 +1,92 @@
+import { ref, readonly } from 'vue'
+
+/**
+ * Push-to-talk recorder wrapping the browser MediaRecorder API.
+ *
+ * Usage:
+ *   const { recording, error, isSupported, startRecording, stopRecording } = useVoiceRecorder()
+ *   await startRecording()
+ *   const blob = await stopRecording()   // resolves with the recorded audio Blob
+ */
+export function useVoiceRecorder() {
+  const recording = ref(false)
+  const error = ref<string | null>(null)
+  const isSupported = typeof MediaRecorder !== 'undefined' && !!navigator.mediaDevices?.getUserMedia
+
+  let mediaRecorder: MediaRecorder | null = null
+  let chunks: Blob[] = []
+  let stream: MediaStream | null = null
+  let resolveStop: ((blob: Blob) => void) | null = null
+  let rejectStop: ((err: Error) => void) | null = null
+
+  async function startRecording(): Promise<void> {
+    error.value = null
+    if (!isSupported) {
+      error.value = 'Audio recording is not supported in this browser'
+      return
+    }
+    if (recording.value) return
+
+    try {
+      stream = await navigator.mediaDevices.getUserMedia({ audio: true })
+    } catch (e) {
+      error.value = 'Microphone access denied'
+      return
+    }
+
+    chunks = []
+    const mimeType = MediaRecorder.isTypeSupported('audio/webm;codecs=opus')
+      ? 'audio/webm;codecs=opus'
+      : MediaRecorder.isTypeSupported('audio/webm')
+        ? 'audio/webm'
+        : ''
+
+    mediaRecorder = mimeType ? new MediaRecorder(stream, { mimeType }) : new MediaRecorder(stream)
+
+    mediaRecorder.ondataavailable = (e) => {
+      if (e.data.size > 0) chunks.push(e.data)
+    }
+
+    mediaRecorder.onstop = () => {
+      const blob = new Blob(chunks, { type: mediaRecorder?.mimeType ?? 'audio/webm' })
+      chunks = []
+      stream?.getTracks().forEach((t) => t.stop())
+      stream = null
+      recording.value = false
+      resolveStop?.(blob)
+      resolveStop = null
+      rejectStop = null
+    }
+
+    mediaRecorder.onerror = () => {
+      recording.value = false
+      error.value = 'Recording error'
+      rejectStop?.(new Error('MediaRecorder error'))
+      resolveStop = null
+      rejectStop = null
+    }
+
+    mediaRecorder.start(100) // collect in 100ms chunks
+    recording.value = true
+  }
+
+  function stopRecording(): Promise<Blob> {
+    return new Promise((resolve, reject) => {
+      if (!mediaRecorder || !recording.value) {
+        reject(new Error('Not recording'))
+        return
+      }
+      resolveStop = resolve
+      rejectStop = reject
+      mediaRecorder.stop()
+    })
+  }
+
+  return {
+    recording: readonly(recording),
+    error: readonly(error),
+    isSupported,
+    startRecording,
+    stopRecording,
+  }
+}