Improving Offline Reinforcement Learning with Inaccurate Simulators