* Optimized RNA property lookups and path resolving, still can be
much better, but now the 1000 IPO example on bf-taskforce25
runs at reasonable speed.
* Also an optimization in the depsgraph when dealing with many
objects, this was actually also a bottleneck here.