Guided Data Augmentation for Offline Reinforcement Learning and Imitation Learning