Non-decreasing Quantile Function Network with Efficient Exploration for Distributional Reinforcement Learning