[CAGE] More rigorous testing for constant tables across threads

[allison]

With the merge of the pcc_reapply branch, test 13 in t/pmc/threads.t will have 3 sections commented out (from r41939). These sections depend on a sub in the constants table being the same as the sub stored in the namespace. This is true for the main interpreter, but not always true for a thread.

Need more extensive testing of the constants table in threads. It could be argued that the constants table should be cloned, or that even direct sub access should be doing a lookup in the namespace for accuracy (with a possible hit in speed).