PR# 15916 Finalizing routine with agent call on redefined function calls wrong version of function

An expanded class has a routine with an agent constructed and called on a
function `value'.  A descendant class redefines `value'.  If the system is
finalized (with or without inlining) the wrong version of the function is called
(the one in the ancestor).  Frozen code works correctly.

Added eweasel test final082 for this bug.

This seems like a pretty serious bug since it won't necessarily be obvious
that the finalized code is wrong (no compiler crash or C compilation failure).
I hope it can be fixed for 6.4.

Run eweasel test final082.

Apologies, this was not fix.

The last full eweasel report on Aug-13 reports this is now fixed.

I couldn't find any "replication" options in the config file, but changing
the inheritance in TEST to "inherit {NONE} TEST" causes the test to pass.
So it sounds like the real problem here is that the compiler should report
a validity error but currently does not.

There are other things we need to finish for the 6.4 release and I doubt we will have time to fix this bug. However, if I recall the intent of what we wanted to do in the standard, the code should be rejected by the compiler since `expanded' actually means frozen (nothing conforms to the expanded entity) and thus you should use `non-conforming' inheritance instead. In which case, it is possible that we actually generate the proper code if the `replication' options is enabled.