Resolve automatic label name detection when single label provided

[ndonthi]

What does this PR do?

Users frequently encounter a confusing KeyError when setting label_names=['label'] in TrainingArguments, leading to errors like:

KeyError: "The `metric_for_best_model` training argument is set to 'eval_f1', which is not found in the evaluation metrics. The available evaluation metrics are: []."

The label_names parameter is designed for advanced multi-label scenarios, but users often mistakenly use label_names=['label'] thinking they need to specify their label column. This breaks the automatic "label" → "labels" conversion that happens during data collation which causes loss calculation to fail silently.

This PR introduces validation for this edge case and fixes the issue in the Trainer to use automatic label detection. If it's preferred, I can make this change where the automatic label detection happens as well!

Implements a code fix to #28530 #40217 and #32373 on huggingface forum

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.
with @
@zach-huggingface, @muellerzr @adrienchaton

[Rocketknight1]

cc @SunMarc

[ndonthi]

checking in on this! @SunMarc @Rocketknight1

let me know if there's another preferred solution :)

[SunMarc]

Thanks for digging ! Sorry for the late review, I missed the notification ... Can you add a test also ? One question I have is how did you get the error in the issue ?