adjust beam search early stopping to any criteria as opposed to all

[secrettoad]

This PR addresses #34843, adjusting the requirement to early stop during generation via beam search to ANY of the supplied criteria as opposed to ALL

Fixes #34843

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?

@Rocketknight1 @gante

[secrettoad]

I'll take a look at the failing tests

[secrettoad]

PyTorch and Tensorflow equivalence test that failed on previous build seems to be non-deterministic, or at the least not reproducible or related to this change. Passing now

[Rocketknight1]

cc @gante @zucchini-nlp

[zucchini-nlp]

For beam search I'd recommend to wait for gante as he had some plans for general refactoring, he'll be back later next week.

[secrettoad]

@gante this seems like an easy fix to the bug mentioned. what do you think?

[gante]

This is actually not the fix -- we might hit the stopping criteria with one of the batch items, but the other batch items would not be finished. The change would also break the case when early_stopping=False

Have a look at #34843 for related discussion, and at #35802 for the proper fix :)