Allow loading whisper models from custom path

[FireFragment]

Current version auto-downloads models to ~/.cache/huggingface/hub/ and tries to load them from there. While convenient, it might not be ideal for every use case. Is it possible to load model from custom path, without auto-downloading anything? Thanks :)

[farzd]

Would be interested in this too

[CodersCreative]

Thanks, I'll look into the issue. However recently I've been planning to change this crate to no longer rely on python and instead use bindings to ctranslate2.