The Unfairness of Fair Machine Learning: Levelling down and strict egalitarianism by default