SlideVQA: A Dataset for Document Visual Question Answering on Multiple Images