Potentially make empirical moments independent of simul_replic

We currently rely on the last simulated series for computing moments. If simul_replic is increased, the moments will therefore change. I would propose to output the first instead of the last run in simult.m. That way, the results would always be the same. Of course, this would entail some loss of backward compatibility. I would accept this, because in principle any replication for the moments should do and the manual states that simul_replic is not used for moment computation. Related to https://forum.dynare.org/t/impact-of-simul-replic-option-on-empirical-moments/10553/4