Allow for Private HF_Lora URLs with Query Params for Auth

[Averylamp]

Please let me know if you have any PR requests or things you'd like to be different.

This change allows a user to pass a custom hf_access_token with the hf_lora url in order to load a custom lora from a private repo.

It follows the existing url path match ^https?://huggingface.co and extracts query parameters to pass directly to pipe.load_lora_weights This allows

https://huggingface.co/<HF_SLUG>/<WEIGHTS_NAME>
https://huggingface.co/<HF_SLUG>?token=<INSERT_HF_ACCESS_TOKEN>&weight_name=<WEIGHTS_NAME>
https://huggingface.co/<HF_SLUG>/<WEIGHTS_NAME>?token=<INSERT_HF_ACCESS_TOKEN>

And other loading parameters to be specified keeping backwards compatibility.

Figured using the existing huggingface url match and url parsing would be a flexible way to pass anything to the load statement without changing anything else, but let me know if you'd like it done a different way

[Averylamp]

FYI will test it with a private deployment to make sure it works, but threw up a PR to get a first read

[gregorym]

@Averylamp Were you able to test it?

[Averylamp]

I did test it awhile back, but then decided that I'd rather use an fp8 compatible lora solution. Saw this outstanding PR replicate/cog-flux#24 and then didn't continue for this one because it seems that it will be duplicate work after cog-flux is updated.
I'd love to get an FP8 Flux Dev Lora on H100s solution to use, but am not sure which repo is correct or what the timeline would be on it.