Only import torch.distributed if it is available

[GaetanLepage]

What does this PR do?

torch.distributed is sometimes unavailable (on some platforms such as darwin).
It leads to transformers.models.auto.modeling_auto to fail to import with:

ModuleNotFoundError: No module named 'torch._C._distributed_c10d'; 'torch._C' is not a package

I propose to implement a similar logic as in huggingface/accelerate#2121 to load torch.distributed conditionally on its availability.
I was able to test the patch in nixpkgs successfully.

Fixes #35129

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[Rocketknight1]

cc @kwen2501 @ArthurZucker - I believe this is caused by the imports added in #34184

[GaetanLepage]

cc @kwen2501 @ArthurZucker - I believe this is caused by the imports added in #34184

Yes, indeed. We vendored a simple patch in nixpkgs: https://github.com/NixOS/nixpkgs/pull/362768/files.
It simply skip the import on darwin and this is good enough for us now.
All the downstream packages depending on transformers seem to work fine.

[ArthurZucker]

LGTM I'll add it to the patch!

[kwen2501]

Looks good, thanks!