Defending Against Backdoor Attacks via Alignment Checking in Model-Contrastive Federated Learning