Reinforcement Learning in Large Population Models : A Continuity Equation Approach ∗