column statistics.

A lot of time was spent in the calculation of the frequencies of values.
For this QMap was used and the sorting in this container is expensive.
The performance of unordered_map is much better. For 10 Mil random doubles the
calculation of the statistics went down from 15.5s to 8.5s on i7-8700.

The next step would be to move out the calculation of the frequencies and
the sorting of the data as well as all dependent calculations to new
threads.