Learning Coordinated Bimanual Manipulation Policies using State Diffusion and Inverse Dynamics Models