Add length-normalized DPO

[hugoabonizio]

What does this PR do?

This PR adds the length-normalized DPO used in Tulu-3 and introduced in the SimPO paper under dpo_norm loss type. This implementation and naming were inspired by the Open-Instruct implementation [1] [2].

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? 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.
• 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.

[qgallouedec]

Hey, thanks for contributing!

Is it really a loss type? It seems to me that it can be combined with any loss type, no?

What about having a new arg in DPOConfig? maybe length_normalize?

Also, I'd add a test for this