Beyond the Universal Law of Robustness: Sharper Laws for Random Features and Neural Tangent Kernels