Learning to Factor Policies and Action-Value Functions: Factored Action Space Representations for Deep Reinforcement learning