Gradient descent follows the regularization path for general losses