feat: define GridMind-RL environment data models and task structures

Add Task 4 instruction following, Curriculum Manager for self-improvement, and world modeling simulation

fix: clamp scores after rounding and ensure all sub-scores are clamped

fix: clamp all scores to open interval (0, 1) to meet validator requirements

feat: add baseline scores JSON, inference script, and update Dockerfile for improved project structure

Initial commit