Be more strict with checking for stat

[scarlehoff]

I think this is a bug although only a bunch of datasets were affected:

"ATLAS_Z0_8TEV_HIMASS_M-Y"
"ATLAS_1JET_7TEV_PTY-R06"
"ATLAS_1JET_7TEV_PTY-R04"
"ATLAS_1JET_8TEV_R06_PTY"

I've checked the chi2 of the first one and it doesn't change, but perhaps it would've had an effect on the shifts.
Basically before "stat" was treated separately. For the dataset that have the statistical uncertainty with another name that is not a problem (they are just an uncertainty) the problem is the ones that had "stat" as the starting of the name but were not actually "stat" (in its legacy meaning) which is the problem that @andrpie saw.

Now it checks for stat at the beginning of the word (just like before) but they are only accepted if they are also UNCORR and ADD. Hopefully this doesn't change the behaviour of old datasets that might rely on that and will also not introduce silent (or not that silent) bugs like in #2418

[scarlehoff]

@enocera sorry, I didn't ask you explicitly. Are you ok with having this check, should I merge this?

[enocera]

Please proceed and merge.