Generative hypergraph clustering: from blockmodels to modularity