Experimentally realizing efficient quantum control with reinforcement learning