refactor: Reward clamping in graders

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

Structure Changed, files reviewed