document Hugging Face authentication for local inference

[Nianze Wu (wunianze666-netizen)]

Summary

This PR adds a short Hugging Face authentication note to the local setup section.

Motivation

While reviewing the local inference instructions, I noticed that inference.py downloads weights from the Hugging Face Hub via hf_hub_download(), but the README currently does not mention that users may need to accept the model terms and authenticate locally first.

This can lead to a 401 Unauthorized error on the first local inference run, as reported in #208.

Changes

• add a small Model access note under local installation
• point users to the model page to accept the terms if needed
• document a minimal hf auth login flow before running local inference

Testing

• Not run: README-only change.

Notes

This is intended to be a small documentation fix for first-run setup friction.