Are Language-and-Vision Transformers Sensitive to Discourse? A Case Study of ViLBERT