Re: Temporary tables versus wraparound... again

Here's a rebased patch. I split the test into a separate patch that I
would lean to dropping. But at least it applies now.

I did look into pg_stat_get_backend_pid() and I guess it would work
but going through the stats mechanism does seem like going the long
way around since we're already looking at the backendId info directly
here, we just weren't grabbing the pid.

I did make a small change, I renamed the checkTempNamespaceStatus()
function to checkTempNamespaceStatusAndPid(). It seems unlikely there
are any external consumers of this function (the only internal
consumer is autovacuum.c). But just in case I renamed it to protect
against any external modules failing from the added parameter.