Leveraging Reinforcement Learning and Large Language Models for Code Optimization