Trainer with Iterable Dataset

[j-rossi-nl]

What does this PR do?

Fixes #5990
Follows #5995 (closed because stale).
Merges all commits from master.

• accomodate iterable dataset without predefined length
• set it as 1 use case: provide max_steps, and NO num_epochs
• Is a merge of master and PR 5995

• This PR fixes a typo or improves the docs (you can dimiss 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 the 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?

@sgugger

[sgugger]

Thanks for your PR. There are a few things I'd like to adjust, on the error messages especially. Or the comments in the tests.

In general, please limit your PR to the functionality you are adding. We can discuss unrelated changes to other parts of the code in separate PRs but it makes reviewing less efficient if you try to group them together.

[sgugger]

Thanks for doing most of the items in the review. I still disagree with your changes in the test file, which makes reusing the RegressionDataset for other frameworks impossible (and is unrelated to the feature you are trying to implement).

[LysandreJik]

LGTM

[sgugger]

Thanks for your contribution!