Model-based Hierarchical Average-reward Reinforcement Learning