RL-Trained Kernel Models

Use this note to compare RL-trained CUDA and Triton generation systems by model, data, verifier, reward, benchmark harness, and evaluation split.