fix: clamp all rewards and scores to [0.10, 0.90]

fix: enforce reward bounds (0.01–0.99) and 2 decimal precision across grader, env, and inference

feat: implement WhyDidItFailState for full OpenEnv state compliance

feat: init the client

refactor: WhyDidItFailAction and WhyDidItFailObservation classes

Initial commit