fix: catch HfHubHTTPError in safetensors auto_conversion thread

[weiguangli-io]

What does this PR do?

Fixes the noisy HfHubHTTPError exception output that appears when loading a transformer model from a repository that has discussions disabled.

Root cause

The previous_pr() function in safetensors_conversion.py calls get_repo_discussions() which raises HfHubHTTPError (403 Forbidden) when a repo has discussions disabled. This exception propagated up through the background Thread-auto_conversion thread, producing a scary traceback during normal model loading.

Changes

1. src/transformers/safetensors_conversion.py: Wrap the get_repo_discussions() iteration in previous_pr() with a try/except HfHubHTTPError block, logging a debug-level message instead of propagating the exception.

2. src/transformers/modeling_utils.py: Change ignore_errors_during_conversion from False to True for the background auto-conversion thread, since this thread is a best-effort optimization for future loads and should never produce user-visible error output.

Fixes #44403

Before

>>> from transformers import AutoModel
>>> bert_model = AutoModel.from_pretrained('hfl/chinese-electra-180g-large-discriminator')
...
Exception in Thread-auto_conversion:
Traceback (most recent call last):
  ...
huggingface_hub.errors.HfHubHTTPError: 403 Forbidden: Discussions are disabled for this repo.

After

>>> from transformers import AutoModel
>>> bert_model = AutoModel.from_pretrained('hfl/chinese-electra-180g-large-discriminator')
>>> # No exception output

🤖 Generated with Claude Code

[Rocketknight1]

Opening PRs to random issues when people are still discussing the issue is not actually helpful!