Major Description
Currently, when we update the Rdn index, we update it many times (one time per RDN in the entry's DN), and we store a new version of each RDN with an updated count of children and descendant.
This is done by dropping and adding each RDN, as we don't have a replace method. This is extermelly costly, and we can save a lot of CPU with such a replace method (the RDN index update cost one third of the global cost of an addition, we could save around half of this, so this would give us a 16% speed improvement for updates.)