Extend EJBTHREE-1116 so that in-flight transactions, specially those initiated remotely, are allowed to
finish and no new transactions are allowed.

EJBTHREE-1116 just blocks the shutdown until in-flight invocations complete, not until transactions
complete. So a tx that's initiated on the client side that spans multiple invocations can see the node
disappear in the middle. Our clustered EJB proxies don't allow failover in the middle of a transaction,
so that means the client sees a failure.

EJBTHREE-1116, consider the following session bean method:

@TransactionAttribute(TransactionAttributeType.REQUIRES_NEW)
public void addUser(User user) {
em.persist(user);
}

If the node is shutdown after a client has called addUser(), the transaction will indeed complete before
the shutdown commences, but only as a side effect of waiting for the invocation to complete - the
BlockContainerShutdownInterceptor sits in the front of the interceptor chain. Subsequent requests to
addUser(...) (i.e. new transactions) will fail once shutdown is initiated.

EJBTHREE-1116 doesn't cover transactions initiated from the client - nor invocations from one ejb to another
(e.g. if addUser() tries to invoke a method on another ejb after shutdown is initiated).