Causal Information Prioritization for Efficient Reinforcement Learning