ConVQG: Contrastive Visual Question Generation with Multimodal Guidance