Using the flag `savepseudodata` doesn't save the same data that was fitted

Created by: scarlehoff

From what I can see for the fit make_replica is called separately for each group, instead when it has to save it 1) it generates it again -which takes time- but also make_replica is called only once, and that messes up for sure with the seeding and maybe with something else.

Basically, both numbers only agree if just one group is used (so for instance only ATLAS data) since in both cases make_replica is called just once.

I think I have a solution for it which I will push when I'm done with making it as minimal as possible.

The training/validation mask seems to be correct instead so not everything is lost!

/cc @J-M-Moore @Zaharid @siranipour @RoyStegeman