WIP: RLOOV2

[mnoukhov]

What does this PR do?

Following #2567, introduces an RLOOv2 that

• follows the structure of OnlineDPOTrainer to leverage Trainer's train and training_step
• adds new GRPOTrainer reward_func so it can be used for reasoning / math
• follows SFTTrainer to automatically preprocess the dataset

and particularly useful for open r1 it follows the current RLOOTrainer

• allows for generating multiple minibatches of data and then training for multiple "ppo epochs" on that data , which doesn't seem to be present in GRPOTrainer

Overall, it updates RLOOTrainer to have all the benefits of Trainer, using the latest format in GRPOTrainer, and maintains all its functionality

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?
• [ x] Was this discussed/approved via a GitHub issue? Please add a 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.