Feature: add FunASR/SenseVoice as optional local ASR provider

[LauraGPT]

Hi, OpenLess's “hold a key, speak, release, polished text appears” workflow is a good fit for pluggable local ASR backends.

Would you consider adding FunASR/SenseVoice as an optional speech-to-text provider?

Why it may be useful:

• SenseVoice supports multilingual ASR with built-in punctuation and rich labels.
• FunASR/Paraformer are strong for Chinese/Mandarin dictation and include VAD/timestamp/punctuation tooling.
• It can run locally, which fits the voice-input/privacy use case.
• It can be integrated as a local HTTP/Python sidecar first, avoiding a deep native integration in the desktop app.

Possible integration path:

1. Add a funasr provider option in the ASR provider settings.
2. Send each recorded utterance to a local FunASR/SenseVoice endpoint.
3. Feed the returned text into the existing LLM polishing / cursor insertion flow.

References:

• FunASR: https://github.com/modelscope/FunASR
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice

Happy to help with a small adapter proposal if this fits the roadmap.