feat: sovereign voice loop — timmy voice command

Adds fully local listen-think-speak voice interface.
STT: Whisper, LLM: Ollama, TTS: Piper. No cloud, no network.

- src/timmy/voice_loop.py: VoiceLoop with VAD, Whisper, Piper
- src/timmy/cli.py: new voice command
- pyproject.toml: voice extras updated
- 20 new tests

src/timmy/cli.py

@@ -248,5 +248,37 @@ def down():
     subprocess.run(["docker", "compose", "down"], check=True)
 
 
+@app.command()
+def voice(
+    whisper_model: str = typer.Option(
+        "base.en", "--whisper", "-w", help="Whisper model: tiny.en, base.en, small.en, medium.en"
+    ),
+    use_say: bool = typer.Option(False, "--say", help="Use macOS `say` instead of Piper TTS"),
+    threshold: float = typer.Option(
+        0.015, "--threshold", "-t", help="Mic silence threshold (RMS). Lower = more sensitive."
+    ),
+    silence: float = typer.Option(1.5, "--silence", help="Seconds of silence to end recording"),
+    backend: str | None = _BACKEND_OPTION,
+    model_size: str | None = _MODEL_SIZE_OPTION,
+):
+    """Start the sovereign voice loop — listen, think, speak.
+
+    Everything runs locally: Whisper for STT, Ollama for LLM, Piper for TTS.
+    No cloud, no network calls, no microphone data leaves your machine.
+    """
+    from timmy.voice_loop import VoiceConfig, VoiceLoop
+
+    config = VoiceConfig(
+        whisper_model=whisper_model,
+        use_say_fallback=use_say,
+        silence_threshold=threshold,
+        silence_duration=silence,
+        backend=backend,
+        model_size=model_size,
+    )
+    loop = VoiceLoop(config=config)
+    loop.run()
+
+
 def main():
     app()