When and How to Fool Explainable Models (and Humans) with Adversarial Examples