MMHQA-ICL: Multimodal In-context Learning for Hybrid Question Answering over Text, Tables and Images