Fine-Tuning Diffusion-Based Recommender Systems via Reinforcement Learning with Reward Function Optimization