Allow RAG to output decoder cross-attentions

[dblakely]

What does this PR do?

This PR makes RAG output the generator model's decoder cross-attentions when output_attentions=True.

Motivation and context: before this PR, RAG's output objects had attributes for the generator's encoder self-attentions and decoder self-attentions, but no option for the encoder-decoder cross-attentions. So this simply allows cross-attentions to be extracted, as well as fixing a small bug where output_attentions wasn't being passed into the generator.

Fixes #9468

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. Yes - Have RAG return generator cross-attentions when output_attentions=True #9468
• 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? - I don't believe any new tests are necessary. Existing tests pass.

Who can review?

@patrickvonplaten, @lhoestq

[lhoestq]

Thanks ! Looks good to me :)

The suggestions below should help you fix the CI.
I think it just comes from a line length issue

[dblakely]

@lhoestq Thanks for the suggestions! All the CI checks pass now.

[lhoestq]

Nice !
Also pinging @patrickvonplaten just to make sure

[patrickvonplaten]

Super! Thanks for you contribution