Head-to-head comparison of clustering methods for heterogeneous data: a simulation-driven benchmark