Improve robustness of background normalisation in CP fit model

[Thomas Latham]

Suggestion from Jeremy Dalseno to improve robustness of normalisation of background model in CP fit model. Instead of doing:

0.5 x (1-q*A_CP) [ (1+q)/2 Q / int(Q) + (1-q)/2 Qbar / int(Qbar)

do

[ (1+q)/2 Q + (1-q)/2 Qbar ] / [ (1-A_CP)/2 int(Q) + (1+A_CP)/2 int(Qbar) ]

which is more robust against either int(Q) or int(Qbar) being 0.

[Thomas Latham]

Looking again at this I don't believe it will work since the integrals of Q and Qbar are arbitrary (unlike for the signal where the integrals of the |A|^{2 for each flavour have meaning). If the histograms supplied as such that Q and Qbar have the same integral then the A_CP parameter drops out.}