Learning Diverse Behaviors: Diffusion Meets RL