Bug 2538

The tests/Portable_Interceptor/Collocated/Service_Context_Manipulation test
fails on a lot of platforms. I was conserned that this was due to some of the
new features, but it isn't. 

The problem is that there is a multithreaded application which contains a
collocated client &  server, the client invokes a method on the server which in
turn calls orb->shutdown(0).

Ordinarily this causes no problems, as the shutdown call happens while the
server is handling the request, so orb->run() doesn't actually return until the
upcall completes.

In this case, the request is actually handled in the caller's thread, so when
shutdown is called, the server's orb->run() exits immediately. In this test, the
server thread calls orb->destroy(), which destroys the server interceptors. This
then causes a segv if the client thread hasn't passed back through the reply
interceptor code in time. Even though the operation triggering the shutdown is a
oneway, since it is in the caller's thread, it still has to return, triggering
the interceptors.

For Collocated/Service_Context_Manipulation, I've made the test not crash by
putting a slight delay after the server's orb->run() returns and before the orb
is destroyed. However, I think that the server orb should really be aware that
an upcall is in progress, even though it is a collocated call, and so orb->run
should not return until that call is done.