Benchmarking Multimodal RAG through a Chart-based Document Question-Answering Generation Framework