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

fix: reward scores are updated to be between 0 and 1

chore: clean up all the unnecessary comments

fix: normalize underfitting gradient norms and guard vague-answer penalty

feat: updating the evidence scoring function

feat: adding steps count logic to encourage the agent explore more

feat: max step limit

feat: fix suggestion is required and not providing fix causes penalty

feat: ordering_bonus function implementation

feat(grade): inspected is upgraded to inspected_order. It rewards steps taken in order

feat: grade function refactor and additional features

feat: init graders and implement grade_easy() in env