feat: acoustic shush detection for instant TTS interruption

[isair]

Summary

• Add a spectral fricative detector (ShushDetector) that runs on raw audio frames during TTS playback, detecting "shhh" sounds in ~300ms without waiting for Whisper transcription
• Analyses each 20ms frame's frequency profile — shush has high energy in 2–8 kHz with little below 500 Hz (no harmonics); requires 15 consecutive frames to trigger, avoiding false positives from brief sibilants in TTS output
• Reduces TTS stop latency from ~1.5–3s (Whisper pipeline) to ~300ms (direct spectral check)

Test plan

• Run unit tests: pytest tests/test_shush_detector.py -v
• Manual test: start Jarvis, trigger a long TTS response, then make a sustained "shhh" sound — TTS should stop within ~300ms
• Verify brief "s"/"sh" sounds in normal speech or TTS output don't trigger false positives
• Test with shush_detection_enabled: false in config to confirm it can be disabled
• Tune shush_consecutive_frames and shush_energy_floor if needed for different mic setups

🤖 Generated with Claude Code