[WIP] GRPO-inspired Online DPO refactor

[d-tiapkin]

What does this PR do?

Large refactoring of Online DPO trainer to make it a subclass of GRPOTrainer and benefit from (a) larger generation batches with vllm and (b) automatically gather all new features to accelerate generation.

The main motivation is that the current implementation does not benefit from vllm integration because the generation batch size is constrained by per_device_train_batch_size, which can be slow for large models and GPU-poor setups.

Current TODOs:

• Support of VLMs;
• Refactor of Nash MD and XPO trainers since they inherit the Online DPO trainer.

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?
• Did you write any new necessary tests?

Who can review?

@kashif @qgallouedec