Rethinking Image Forgery Detection via Contrastive Learning and Unsupervised Clustering