[Feature Request]: Switch whisper to onnx runtime

[thewh1teagle]

Describe the feature

Currently sherpa-onnx supports coreml and cuda.
Once it will support rocm for amd GPUs we'll switch to use it with sherpa-rs bindings.

Tracking issue: k2-fsa/sherpa-onnx#196

[thewh1teagle]

Tasks before:

[ ] (Maybe we'll keep whisper.cpp for stt) Wait for ggerganov/whisper.cpp#2279 merge
[ ] Whisper model format of onnx it too complicated. needs to find a way to pass single file so users can change it
[ ] Fix #167
[ ] Add migrate UI that will delete old ggml file and download new model
[ ] Check that speed is good like whisper.cpp
[ ] Prepare faster-whisper-v2-d3-e3/tree/main in onnx format
[ X ] Implement Max speakers for diarize
[X ] Fix k2-fsa/sherpa-onnx#1085
[ ] Add amd support to sherpa k2-fsa/sherpa-onnx#196
[ ] Check that it return Hebrew correctly
[ ] Check that it works on mac, Linux, Windows