Unlock: Reinforcement Learning Environments and Benchmarks

The standard RL evaluation stack: Gymnasium API, classic control tasks, Atari, MuJoCo, ProcGen, the sim-to-real gap, and why benchmark performance is a poor predictor of real-world RL capability.