API changes in vtkDescriptiveStatistics

• Kurtosis formula had a mistake that is now corrected

• API of this class has changed. Previously, one could turn on and off which version of the kurtosis, skewness and variance was to be computed: the sample version or the population version. This was being done by toggling on and off UnbiasedVariance for the variance, G1Skewness for the skewness, and G2Skewness. This switch is now centralized in the new SampleEstimate boolean. When turned on, the sample version of each statistics is computed. When turned off, the population version is instead. By default, the sample version is computed.

This change is introduced to simplify the interaction with the filter, and to avoid confusion from the user, who used to be able to compute mixed types of statistics, when one shouldn't. Indeed, given an input dataset, it makes no sense to compute the sample skewness while computing the population kurtosis, as this choice should be made globally, depending on what kind of statistics is wanted. Both should be either the sample version, or the population version.

• The documentation is enhanced, and the explicit formulas for each statistics in each case is fully given.

• A zero variance now outputs NaN values for skewness and kurtosis, as it should do, as skewness and kurtosis are undefined under such circumstances.

Adresses paraview/paraview#20854 (closed)