feat: basic vllm support for hf cached models

[axel7083]

What does this PR do?

Quick POC showing integration of VLLM in AI Lab.

VLLM only support .safetensors models. AI Lab only support the GGUF models, this PR show how the amazing @huggingface/hub library can be used to read the downloaded models from the cache and mount them to create an inference server.

I build and push myself an image quay.io/rh-ee-astefani/vllm:cpu-1734105797 this image has been built following the instruction in https://docs.vllm.ai/en/latest/getting_started/cpu-installation.html#quick-start-using-dockerfile

Screenshot / video of UI

What issues does this PR fix or reference?

How to test this PR?

1. install the huggingface cli tool (See Installation)
2. Run huggingface-cli download Qwen/Qwen2.5-0.5B-Instruct (it will download a model)
3. Start/Restart AI Lab, the hugging face models are loaded on startup.
4. Assert Qwen/Qwen2.5-0.5B-Instruct is in the imported section of the models catalog
5. Click on Create Model service
   
6. Click on Create Service
   
7. assert service is created
   
8. Open Podman Desktop containers page
9. assert your cpu-vllm container is running
10. Wait for the server to be up and running (can take a few minutes)
11. checks models list
    
12. check basic request