Riemannian Motion Policy Fusion through Learnable Lyapunov Function Reshaping