e_tree_model is apparently used for flat lists as well as parent-child
relationships (trees). Trees are expected to be far smaller than possible flat
lists. Very large flat lists (10,000+ rows or greater) are rendered very slowly
because of the tree computation overhead.
This change figures out whether a flat list or a tree is requested and chooses
the appropriate code to run based on what is requested. Trees run the more
expensive code while flat lists are returned as-is.
In addition, the tree rendering code has been optimized. Optimizations:
* Unchanging tree node ID is set once instead of inside a foreach() loop
* The parent-child query is now sorted by the sort parent ID so that each move
rows to tree nodes iteration doesn't have to run through every remaining row
Fixes: #3062
8a675021ff