Uncertainty-aware Distributional Offline Reinforcement Learning