Toward Automatic Relevance Judgment using Vision--Language Models for Image--Text Retrieval Evaluation