Efficient Model-Based Concave Utility Reinforcement Learning through Greedy Mirror Descent