Membrane Permeating Macrocycles: Design Guidelines from Machine Learning