Add support for SenseVoice model, which is supposedly much better than whisper

[Tamnac]

https://github.com/FunAudioLLM/SenseVoice for reference. It claims to have much better latency for the small model, and better performance on non-english languages, especially chinese/cantonese.

[jart]

The model would need to be implemented in C++. We'd happily welcome contributions doing that, if anyone wants to take up the cause of porting this from Torch to GGML.

[Tamnac]

I saw a project of someone doing that, see here https://github.com/lovemefan/SenseVoice.cpp I even downloaded their model, however it didn't run with whisperfile.