Examining Modality Incongruity in Multimodal Federated Learning for Medical Vision and Language-based Disease Detection