Towards Multi-agent Reinforcement Learning based Traffic Signal Control through Spatio-temporal Hypergraphs