Feature: Add SenseVoice for on-device STT

[LauraGPT]

Motivation

Parlor focuses on on-device realtime multimodal AI. SenseVoice (8K+ stars) from FunASR is well-suited for on-device STT:

Why SenseVoice

• Non-autoregressive: Single forward pass — minimal latency for realtime conversations
• 234M params: Small enough for on-device inference
• 50+ languages with auto detection
• Emotion detection: Built-in — useful for multimodal AI context
• ONNX available: Via Sherpa-ONNX for iOS, Android, embedded devices

References

• FunASR: https://github.com/modelscope/FunASR
• Sherpa-ONNX: https://github.com/k2-fsa/sherpa-onnx