Efficient Distributional Reinforcement Learning with Kullback-Leibler Divergence Regularization