Can Active Sampling Reduce Causal Confusion in Offline Reinforcement Learning?