refactor: Update task configurations and grading logic for improved scoring and consistency

refactor: Reward clamping in graders

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

refactor: Improved grading logic for task 2

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

fix: Update file paths and ensure model loading in PropertyRetriever

Structure Changed, files reviewed