Global Convergence Guarantees for Federated Policy Gradient Methods with Adversaries