Generative hypergraph clustering: From blockmodels to modularity