Curriculum Goal-Conditioned Imitation for Offline Reinforcement Learning