ColPali: Efficient Document Retrieval with Vision Language Models