Guaranteed Optimal Compositional Explanations for Neurons