ColPali: Vision Language Models for Efficient Document Retrieval