Distribution shift, generalization and OOD challenge in offline reinforcement learning: a comprehensive survey | Synapse