Evaluating Gradient Inversion Attacks and Defenses in Federated Learning