Rethink MPTT threshold between update and rebuild
#5399 (closed) changed from len(ids) < count / 4
to len(ids) < 2
. But it seems there is a flaw in the reasoning because we count "2 write on average ½ all nodes" but new nodes are mainly added at the end of its branch. For example when adding new main records, their update will update only themselves. It is difficult to get a good average of the number of nodes updated.
So I'm wondering if the previous threshold was not better if we add cache to the counting (as we only need an approximation).