provide offline option for llama_download_file

[codefromthecrypt]

Right now, if you are on a plane, or otherwise offline, and you invoke something that calls llama_download_file, you will get a failure verifying the already downloaded model.

I would love an offline arg, so that I can work with the last version for any reason including network failure.

For example, I would like this to not fail if it has a local file, and don't mind passing another arg to clarify

$ llama-server --log-disable \
  --hf-repo Qwen/Qwen2-0.5B-Instruct-GGUF \
  --hf-file qwen2-0_5b-instruct-q5_k_m.gguf