only validate hf user token on rank 0

[winglian]

When training with multi-node, it's easy to hit rate limits, this ensures that we only check the token on rank 0 which should be sufficient to fail the process if it is found to be invalid.

[rank3]: Traceback (most recent call last):                                                                                                                                                                                                    [rank3]:   File "/root/miniconda3/envs/py3.11/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 1639, in whoami                                                                                                                    
[rank3]:     hf_raise_for_status(r)                                                                                                                                                                                                            [rank3]:   File "/root/miniconda3/envs/py3.11/lib/python3.11/site-packages/huggingface_hub/utils/_http.py", line 477, in hf_raise_for_status                                                                                                   
[rank3]:     raise _format(HfHubHTTPError, str(e), response) from e                                                                                                                                                                            [rank3]: huggingface_hub.errors.HfHubHTTPError: 504 Server Error: Gateway Time-out for url: https://huggingface.co/api/whoami-v2