We use the renderer to create the reference for the clip test by
applying the same clip node to the reference image instead of the
tested node.
This is somewhat suboptimal because they run very similar codepaths, but
this method works with high bit depth content and different colorstates