Layout-Aware Information Extraction for Document-Grounded Dialogue: Dataset, Method and Demonstration