Offline Reinforcement Learning from Datasets with Structured Non-Stationarity