Group-Adaptive Adversarial Learning for Robust Fake News Detection Against Malicious Comments