Formal Verification of the Adversarial Robustness Property of Deep Neural Networks Through Dimension Reduction Heuristics, Refutation-Based Abstraction, and Partitioning