Fix skipped gradient update of the last step when using gradient accumulation.

[richardSHkim]

What does this PR do?

This PR fixes skipped gradient update of the last step when using gradient accumulation in trainer.py.

I'm not sure what is the meaning of 'step is always smaller than gradient_accumulation_steps', but steps_in_epoch <= args.gradient_accumulation_steps looks like always False.
Thus, the update of the last iteration is skipped when gradient_accumulation_step > 1.

This also leads to different total iterations (-1 iteration per epoch).

It might be negligible, and the skipped gradients are used in next iteration, but I believe the gradient update behavior should be the same whether or not using gradient accumulation if the total batch size is the same.

I have tested with example script and the comparison results are here.

accelerator.accumulate() context manager now can handle the last step of data loader, so I'm trying to use it.

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?

@muellerzr @pacman100

[richardSHkim]

I found this implementation was intended in PR#24415, so close this PR.