Fixed the discrimination script to correctly account for Poisson statistics for the overall number of counts. After this, the loglikelihood is a much better discriminator than debinning. However, the debinning calculation effectively ignores the total number of counts. Perhaps an improvement to the debinning calculation which involves Poisson statistics can make it competitive.

Also, I fixed the nKr calculations to use (...).np.sum() instead of
reduce(op.add, ...).

Changes to be committed:
modified: debinningCore/debinning.py
modified: debinningMath/orderStatistics.py
modified: examples/discriminationMC.py