Towards Empowerment Gain through Causal Structure Learning in Model-Based RL