Exploring Diverse In-Context Configurations for Image Captioning Xu Y ang