Steering Your Diffusion Policy with Latent Space Reinforcement Learning