AI Meet DeepCache: A Simple and Effective Acceleration Algorithm for Dynamically Compressing Diffusion Models during Runtime