Add support for safetensors (non-GGUF) model format

[sallyom]

Instead of a single model file, ramalama will need to mount and pull a directory. It would be really nice to have ramalama pull the/model --format safetensors or something like that

[bmahabirbu]

Great idea, if we're using vllm (or others that support safe tensor) as the runtime we can mount the folder as is. If we are using llama.cpp we can convert the safetensors to a gguf then mount the file as usual!

[ericcurtin]

I agree we should implement this. I would say even drop the "--format safetensors", I think it would not be too hard to automatically detect a certain model is safetensors.

[rhatdan]

SGTM