Order Optimal One-Shot Distributed Learning