Securing NextG Systems against Poisoning Attacks on Federated Learning: A Game-Theoretic Solution