Beyond the CLS Token: Image Reranking using Pretrained Vision Transformers