* Undo 'any' inference propagation
Removing this only changes one test slightly, and fixes JQuery types,
which rely on the old method of inference.
* Add jquery regression test and update baselines
* Restore any inference propagation to wildcard only