Gender Biases in Automatic Evaluation Metrics for Image Captioning