On the Impossibility of Complete Interpretability in Neural Networks: A Measure-Theoretic Analysis | Synapse