Manipulating the Byzantine: Optimizing Model Poisoning Attacks and Defenses for Federated Learning | Synapse