refactor: Reward clamping in graders

refactor: Task3 reward model changed, agent adjusted for new model

refactor: Update grading logic and submission handling across tasks for improved accuracy and consistency

Structure Changed, files reviewed