Sample and Computationally Efficient Robust Learning of Gaussian Single-Index Models