llhr weights need negative sign for maximisation

[APN-Pucky]

We noticed the signal regions were not combined optimally otherwise (path_finder maximizes the weights).

[jackaraz]

Thanks @APN-Pucky; the typo is actually coming from the initial computation of $\chi^2$. llhr should not have been negative in the first place, and it should have been computed with respect to apriori statistical model as done in the original paper where we proposed the method. I fixed it in #17 and will release it once the branch is validated. Thanks again for pointing this out!