Joint Extraction Matters: Prompt-Based Visual Question Answering for Multi-Field Document Information Extraction