chore: update doc string

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

fix: tighten label rules for underfitting, overfitting, and vanishing gradients

fix: add reasoning in Action class

feat: define WhyDidItFailAction and WhyDidItFailObservation models with typed fields and descriptions

Initial commit