Blender Git Loki

Depsgraph: Fix frash with iTaSC solver

This commit reverts part of a fix for T33275, but things are:

- I can not reproduce the original issue at all, so doesn't seem to
cause any regressions.

- It is really bad idea to do delayed initialization in the threaded
environment, it's a straight way to some nasty issues.

- We can't do things like this anyway because we go more granular,
meaning such a delayed initialization will fail in the case of
having several IK solvers (unless they properly accommodate to
changed bone head).

- Verified the fix with various files from Mango project and all of
them seems to work nice with new depednency graph now (old depsgraph
has some flickering, but it's not related on DEG itself, but on
an environment with lots of proxies and threaded evaluation and it
is not a new behavior).