feat(voice): chunked TTS playback for long text

[DannyNs]

Summary

• Split long TTS text into rolling chunks of 2-3 sentences (~150 chars each)
• Pipeline generation + playback: first chunk plays almost immediately while next generates in background
• Configurable chunk threshold (default 100 chars) via dashboard UI and config
• voice_stop cancels instantly between chunks via stop flag
• Falls back gracefully: chunked → single-shot edge-tts → platform native TTS

Reference

Inspired by dyoburon/vibetotext#9 (Python implementation)

Changes

• src/apps/voice/src/services/tts.ts — splitSentences(), groupChunks(), playMp3Blocking(), speakChunked() + updated speak() and stop()
• src/apps/voice/src/services/config-store.ts — Added chunkThreshold to TtsConfig
• src/apps/voice/public/page.js — "Chunk Threshold (chars)" field in dashboard TTS settings

Test plan

• Short text (<100 chars): verify single-shot playback unchanged
• Long text (500+ chars): verify first chunk starts playing within seconds, seamless transitions between chunks
• voice_stop during chunked playback: verify immediate cancellation
• Set chunk threshold to 0 in dashboard: verify all text goes through chunked path
• Set chunk threshold to 5000: verify long text falls back to single-shot
• Verify temp chunk files are cleaned up after playback completes

🤖 Generated with Claude Code