Evolution of Multimodal Question Answering: From Modality-Adaptive Extraction to Unified Language Representation