Provable Fairness for Neural Network Models using Formal Verification