Add Provence model implementation

[matthewhaynesonline]

Add implementation for https://huggingface.co/naver/provence-reranker-debertav3-v1. This is still a WIP, but I wanted to gauge interest before going too far.

Notes

• Provence has a CC Non Commercial license
  • Not sure if that's against candle conventions to implement models with restricted licenses
• Since it's based on DeBERTa-v3, the core implementation is pretty easy (forward) however the original py transformers implementation provides a custom process function (and thus requires remote code)
  • Since by convention the candle models implement forward but not full pipelines, I wasn't sure if adding a process function in the model would be a bad choice.
  • For now it's there, but feature flagged
• I started by looking at the DeBERTa example and the XLM-RoBERTa reranking and tried to go from there.
• There's other versions (multilingual) and an LLM approach, but this particular model was a good fit for a use case of mine, so that's why I've focused on it
• If there's potential for approving this, I can reach out to the original authors on HF to get their feedback as well

Thank you!