Dynamic Angle Selection in X-Ray CT: A Reinforcement Learning Approach to Optimal Stopping