feat: prioritize gguf and safetensors formats for embeddings and LoRAs

[wbruna]

Models available as multiple file types were likely converted from one another, so we should prioritize native, safer and more efficient formats.

[leejet]

I’m not entirely sure whether this is necessary here. Generally, GGUF files include the corresponding quantization format in their filenames, so they differ from the original version.

[wbruna]

My main concern is with the current .pt priority. I agree it doesn't matter that much for .gguf versus .safetensors (as I also don't see a strong reason for preferring .safetensors over .gguf, either; but I can adapt the PR if there is one).