[Plugin] Add livekit-plugins-funasr for self-hosted STT

[LauraGPT]

Implementation status

PR #6176 adds livekit-plugins-funasr as an optional local STT plugin.

The current implementation is intentionally narrow:

• direct local inference with iic/SenseVoiceSmall;
• batch STT.recognize integration using LiveKit audio frames;
• 16-bit PCM conversion and resampling to 16 kHz;
• cleaned transcription text with language detection;
• an asyncio.Lock around shared model inference;
• the livekit-agents[funasr] optional dependency extra and complete Torch/Torchaudio runtime dependencies.

SenseVoiceSmall supports Mandarin, Cantonese, English, Japanese, and Korean. It can also emit speech-emotion and audio-event tags; the plugin returns clean transcription text and does not present nospeech as a language. Model weights follow the license linked from the model card.

Validation

The current PR head has been verified with:

• three focused FunASR plugin tests;
• Ruff, formatting, and mypy across the repository;
• wheel and sdist builds;
• standalone Python 3.10 dependency resolution;
• real SenseVoiceSmall inference through STT.recognize;
• GitHub CI with no failures.

Whisper and other LiveKit STT plugins remain unchanged. Streaming, diarization, and a remote server mode are outside the scope of this first plugin.