Incorporating Spatial Information into Goal-Conditioned Hierarchical Reinforcement Learning via Graph Representations