Kantian Deontology Meets AI Alignment: Towards Morally Robust Fairness Metrics