Goal Discovery with Causal Capacity for Efficient Reinforcement Learning