C++ backend parallelism bug

[jpeoples]

There is a race condition in Filter.cpp. Because the algorithm runs in parallel, accessing the mpFeatureInformationMatrix->at() method is a bug, because it both reads and writes the underlying data, due to the lazy computation of the MI matrix.

It is possible there are other parallelism related bugs I haven't noticed, but I know that in preparing #13 , I had to disable parallelism in the R implementation in order to get the results to match between the two libraries.

I'll submit an issue to the R library as well.