Add separate correlated muR+F variation, fix up/down directions

It might be interesting to try to also include the correlated muR+F variation as a separate nuisance (or we might be asked to do it at some point). Even if don't, it's not much work to produce so we'll already have it in.

Also fixed the directions in the scale variations so that "down" also corresponds to scaling the weight down.

I've also added protections against non-existent theory weights for the alternative ttbb samples (in some cases the branch is there but only contains dummy weights).