[Gradient checkpoining] Correct disabling find_unused_parameters in Trainer when gradient checkpointing is enabled

[patrickvonplaten]

What does this PR do?

Following #13657 this PR makes sure that the Trainer uses the new gradient_checkpointing logic to disable find_unused_parameters argument in DDP.

@sgugger - don't think many people have switched from self.config._gradient_checkpointing to the new API yet, but for those that have previously find_unused_parameters would not have been set to False which then leads to some hard to debug problems like: https://discuss.pytorch.org/t/finding-the-cause-of-runtimeerror-expected-to-mark-a-variable-ready-only-once/124428/5

Not sure if this worth a patch or not

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.

[sgugger]

Thanks a lot for fixing! The check in the Trainer was changed to match the intermediate iteration in the PR for gradient checkpointing and I forgot to adapt it to the last one.