From here to infinity: sparse finite versus Dirichlet process mixtures in model-based clustering