cloned interpreters now actually run and pass all but 55/10386

cloned interpreters now actually run and pass all but 55/10386
subtests; various subtle bugs, new and old, observed when running
cloned interpreters have been fixed

still to do:
|  * dup psig_ptr table
|  * merge PADOP GVs support with "our" SVs (existing PADOPs are too
|    simple-minded and grab one pad entry each, heavily bloating
|    the pad by not avoiding dups)
|  * overloaded constants are not really immutable--they need to
|    be PADOPs
|  * allocator for constants and OPs need to be spelled differently
|    (shared vs interpreter-local allocations)
|  * optree refcounting is still missing locking (macros are in place)
|  * curstackinfo, {mark,scope,save,ret}stack need to be cloned so
|    perl_clone() can be called from within runops*()

p4raw-id: //depot/perl@4553