fix onnx export of speech foundation models

[nikosanto13]

What does this PR do?

This old issue #10004 described an error in the onnx export of wav2vec2-base-960h.
The issue was partially fixed for some of the speech foundation models in this PR: #16004.

This PR:

• applies the same change of expanding the attention mask in all modeling scripts that this occurs (e.g. I came across this when trying to export wavlm-base-plus (modeling_wavlm.py) to onnx),
• applies the same change of expanding the downsampled padding mask in the <Wav2Vec2, Hubert, ...>ForSequenceClassification modules' forward implementation, because the onnx export of these fails similarly, due to the same broadcasting-related error

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?

Tagging @ylacombe, @eustlb (let me know if I should also add someone else)