refactor: Reward clamping in graders

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

refactor: Improved grading logic for task 2

refactor: Update ActionType to include costs and modified grader for task 1

task1, task2 evaluated

task3 reviewed

task2 reviewed, semanticmatcher implemented

Task 3 Implemented

Task 2 added

Inital Commit