Optimal Statistical Guaratees for Adversarially Robust Gaussian Classification