Optimizing Learning Rate Schedules for Iterative Pruning of Deep Neural Networks