The Price of Implicit Bias in Adversarially Robust Generalization