[Feature Extractors] Return attention mask always in int32

[patrickvonplaten]

What does this PR do?

This PR fixes:

tests/test_modeling_tf_hubert.py::TFHubertModelIntegrationTest::test_inference_ctc_robust_batched
tests/test_modeling_tf_wav2vec2.py::TFWav2Vec2ModelIntegrationTest::test_inference_ctc_robust_batched

For some specific use cases the attention mask for feature extractors was returned to be of type bool which broke two tf slow tests. Make sure that it's always int32 or long just like the tokenizers do for text.

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.

[LysandreJik]

LGTM, wdyt @Rocketknight1 ?

[anton-l]

Works like a charm! Since the tensor casting logic is not changed on the TF side of the library, this shouldn't cause any problems down the road.

[Rocketknight1]

LGTM! I think lots of stuff breaks with boolean attention masks, so we definitely want to avoid sending that to the model.