TACO: Enhancing Multimodal In-context Learning via Task Mapping-Guided Sequence Configuration