Motion-Conditioned Diffusion Model for Controllable Video Synthesis