Reporting: make order of validation plots configurable

[kreczko]

• order by some preference (custom sorting)
• order by the largest discrepancy
• segment by "important" and "less important" distributions

[beojan]

There could also be a configurable way to decide if two distributions match. When using floating point variables, numerical instabilities can lead to differences even if the KS test comes out perfect.