Enumerating Safe Regions in Deep Neural Networks with Provable Probabilistic Guarantees