TAPE: Leveraging Agent Topology for Cooperative Multi-Agent Policy Gradient