We can make TCP with bundling on work (timeout 0) if we exchange order
of passUp/passDown of the view for a new member in Gms line 509. Since
bundler is sending messages in bundles it screws up synchronous step
process required for a new member when doing START_FLUSH, view
installation, STOP_FLUSH. We have a thread race here on line 509. In
order to circumvent that we can first pass a view from a join response
up to FLUSH and then pass it down. I am not sure how this affects other
tests.