On the Fairness ROAD: Robust Optimization for Adversarial Debiasing