Towards a Robust Deep Neural Network in Texts: A Survey