Draft: Add GooPdf with statistical background subtraction

[thboettc]

This MR adds GooPdfBkgSub, which is a GooPdf along with an UnbinnedDataSet that can be used for subtracting the background contribution to the likelihood. The UnbinnedDataSet can be used to describe any background contribution that can't be parameterized and incorporated into the underlying GooPdf. Note that this implementation is only valid for un-extended, unbinned NLL fits.

This is based on the method used in AmpTools described here: https://github.com/mashephe/AmpTools/blob/master/AmpTools_User_Guide.pdf.

FYI @cpappenheimer @FlorianReiss @JuanBSLeite

TODO:

• Add an example for testing
• Add python binding
• Check underlying GooPdf and fit data for validity (un-extended fit, unbinned fit data)

Optional add-ons:

• Add support for extended fits
• Add data generation method (could be accomplished by resampling background events)