Planning-Guided Diffusion Policy Learning for Generalizable Contact-Rich Bimanual Manipulation