Add whisper.cpp and thinking model support

[OutisLi]

1. Add whisper.cpp support, which support various hardware architecture, including CPU, NVIDIA GPU, AMD GPU, Intel GPU and so on. Later I will write a doc about how to compile whisper.cpp, however if others can supply a standalone binary distribution will be awesome.
2. Add thinking model support
3. force translate_parallel_num to be 1 when using local model.