[WIP] Refactor attention modules in Bert-based models to use global attention functions

[Marcel256]

What does this PR do?

Refactoring of the Attention Modules in bert-based models to use global attention functions.
Allows support of SDPA and flash attention for Bert models and potentially flex attention in the future

This is a first draft (Feedback appreciated)
Could solve:
#37105
#29129
#27957

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• [x ] 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.