FormNetV2: Multimodal Graph Contrastive Learning for Form Document Information Extraction