fix(wandb): pass fake dataset to avoid exception in trainer (see #34455)

[CezaPasc]

What does this PR do?

This PR passes a fake evaluation dataset for the initialisation of the fake_trainer of WandB.
This is necessary, since the init of the Trainer checks if an eval_strategy is set and raises an error if the eval_dataset is missing.
This issue is documented in #34701.
The fix here is similar to the one in #34455.

Fixes # (issue)

• Pass ["fake"] as eval_dataset during initialisation of fake_trainer for saving purposes of WandB.

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.

@SunMarc

[LysandreJik]

Hello @CezaPasc, I believe this should have been fixed with #34455

[LysandreJik]

Ah, my bad, I see where this one is different. There are a few files which are changed which shouldn't be; do you mind keeping the diff only to the fix here? Thank you!

[CezaPasc]

@LysandreJik
It seems like I got the other diffs from rebasing with the upstream.
I reverted them so that the diff should now only be about the changes in this PR again.

[LysandreJik]

Cool, thank you!