Deep Neural Networks with General Activations: Super-Convergence in Sobolev Norms