Support direct download of models from Hugging Face for deployment

https://github.com/vllm-project/aibrix/blob/main/samples/kvcache/l1cache/vllm.yaml

It seems that models need to be downloaded from storage to local pods. Can we directly download models from Hugging Face to pods like how we deploy large models with vLLM? This would be simpler and more convenient.

Start with samples/kvcache/l1cache/vllm.yaml and trace how models are currently downloaded from storage to local pods. Compare that deployment flow with direct Hugging Face model downloads used by vLLM. Done means the deployment can retrieve a model directly from Hugging Face into its pods and use it successfully.

Written by the indexing model from the issue text.