Towards Minimax Optimal Reinforcement Learning in Factored MDPs