[MaskFormer, Mask2Former] Use einsum where possible

[amyeroberts]

What does this PR do?

einsum operations were removed from mask2former, oneformer and maskformer in #25297 to enable tracing on the models. This was with pytorch 2.0.1

This however, causes a large memory footprint, partially addressed in #25741. However, the models remain very slow.

Testing on recent versions of python >= 2.1 tracing now works for einsum operations.

This PR adds back the previous einsum operations, falling back to the slow but compatible versions when pytorch < 2.1 and the model is being traced.

Fixes #27650

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.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ydshieh]

Nice it works now for torch >= 2.1. Thank you for the update.