EDIT

Currently in TaxonNode.listChildrenOfRecursive 2 different implementations are used to compute the result (first is using Criteria for all TaxonNodes while with comparator solution uses recusive call to children). This is not so nice.

We could better implement a TaxonNodeByParentRankAndNameComparator (and same for name only and natural) which includes comparing the first parents which differ (if parents differ and if not the one is an ancestor of the other - which makes sorting clear -).

This shouldn't be too difficult by using TreeIndex methods.

Another even better solution whould be to generally use the TreeNodeFilter and add parent,rank and name ordering to the taxon node filter.

A general question is if we need to initialize taxon nodes at all (seems to be an overhead for the distribution editor as it does need only data from taxon and below.