Add line numbers to code display, fix clear_flag loop, match submission spec exactly

Fix stdout to [START]/[STEP]/[END] format, use OpenAI client with HF defaults

Add diversity/exploration bonuses, near-miss type check, context truncation

Add 7-task RL env with PBRS, CAMRL curriculum, VL norm, RC-GRPO inference

Accept OPENAI_API_KEY as fallback for HF_TOKEN

Code Review Environment OpenEnv hackathon submission