Event-Oriented Visual Question Answering: The E-VQA Dataset and Benchmark