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