Doesn't work correctly yet! Midway refactoring: removing the possibility
to run anova on multiple covariables, as this would be best implemented
using a) the formula and b) a linear regression model as a first step.
Since we don't yet have formulae and algorithm pipelines there is no
point in implementing the multivariate anova version.

Plus, add Tukey table to raw output.

Adds statmodels in Dockerfile also