The Distributional Reward Critic Architecture for Perturbed-Reward Reinforcement Learning