Wyoming TTS: crash when client specifies a voice — SynthesizeVoice object passed where str expected

[crudiyay]

Bug

agent_cli/server/tts/wyoming_handler.py (v0.101.0, lines 93/95) passes the raw Wyoming SynthesizeVoice object into the model manager:

await self._synthesize_streaming(manager, text, synthesize.voice)
...
await self._synthesize_complete(manager, text, synthesize.voice)

manager.synthesize(text, voice=...) expects str | None, so any client that specifies a voice crashes synthesis:

RuntimeError: argument should be a str or an os.PathLike object where __fspath__ returns a str, not 'SynthesizeVoice'

The handler's exception path then returns empty audio, so the client just hears nothing.

Impact

Home Assistant sends SynthesizeVoice whenever the user picks a voice in the pipeline settings — i.e., voice selection in HA makes TTS silently fail (works only when no voice is chosen). Combined with #592 this makes the Kokoro Wyoming path effectively broken for HA out of the box.

Repro

1. agent-cli server tts --backend kokoro --model am_adam --model bm_george (v0.101.0, macOS arm64, uv tool install)
2. HA Wyoming integration → pipeline TTS = agent-cli-tts, voice = bm_george
3. Trigger any TTS (pipeline response or assist_satellite.announce) → no audio; server log shows the RuntimeError above

Fix

voice = synthesize.voice.name if synthesize.voice else None

(Verified locally — with this one-liner HA voice selection works, including speaker selection via SynthesizeVoice.speaker being ignored gracefully.)