Enable RLHF training for pytorch transformer

[ealt]

Enable Reinforcement Learning from Verifier Rewards (RLVR) training for PyTorch transformer models using the TRL library.

This allows training models with reward signals for arithmetic tasks, configurable via YAML files and runnable similar to existing training scripts, incorporating boxed_answer_reward and correct_answer_reward functions.

---

[cursor]

Cursor Agent can help with this pull request. Just @cursor in comments and I'll start working on changes in this branch.
_{Learn more about Cursor Agents}