Add example of RL training using open-instruct

[olliestanley]

This post-training codebase supports GRPO / RL from verifiable rewards. It would be good to have an example in the repo showing how it can be used with reasoning-gym procedural datasets

https://github.com/allenai/open-instruct

Existing examples (with trl, veRL, unsloth, OpenRLHF) can be used as inspiration