Statistics of Stealing Machine Learning Models: Attacks and Countermeasures for Generative Adversarial Networks