generation/stopping_criteria: short-circuit StoppingCriteriaList when all sequences are done

[GitGlimpse895]

What does this PR do?

StoppingCriteriaList.__call__ previously evaluated every registered criterion
unconditionally on every generation step, even after is_done was already True
for all sequences in the batch. This adds a single if is_done.all(): break guard
inside the loop to skip remaining criteria once the entire batch is finished.

This is semantically safe because OR-ing any further True values into an
all-True tensor cannot change the result. The saving scales with
(number_of_criteria − 1) × remaining_steps_after_completion, and is most
impactful when StopStringCriteria (which runs a full vocab embedding lookup via
F.embedding each call) appears after a cheaper criterion like MaxLengthCriteria.

A new test test_list_criteria_short_circuits_when_all_done is added to verify
the behaviour using a CountingCriteria sentinel.

• I confirm that this is not a pure code agent PR.

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?
• Did you write any new necessary tests?

[Rocketknight1]

Does this not create data-dependent control flow that requires a compiler break?

[GitGlimpse895]

You're right that calling .all() on a tensor and using it in a Python if/break creates data-dependent control flow. Under torch.compile this would cause a graph break. However, StoppingCriteriaList.call is invoked from the Python-level generation loop in utils.py, which itself runs outside any compiled region — the model forward pass is compiled, but the decoding loop that calls stopping criteria is not. So no graph break is introduced in practice. That said, I'm happy to add a # pragma: no compile note or restructure if you'd prefer a different approach.