Plan To Predict: Learning an Uncertainty-Foreseeing Model For Model-Based Reinforcement Learning