Repro - Differentiable Optimization Layers for Guaranteed Fairness in Deep Learning