Behind the scenes: benchmarking sub-grid scales models in global simulations of stellar convective dynamos