Add code-rl recipe with DeepCoder

[Xiuyu-Li]

trafficstars

This PR adds an example recipe for reinforcement learning (RL) to solve competitive programming problems using Tinker and the DeepCoder dataset. The recipe is located at tinker_cookbook/recipes/code_rl. The environment uses sandboxing via Sandbox Fusion for security, without introducing any additional dependencies when running in Docker.

[TieMoulton]

this is pretty neat, i agree

[joschu]

Cool!

[joschu]

Thanks for adding this! We'll review shortly!

[Tiiiger]

Great! I ran the experiment in the README and seems to be working as expected. Merging now