Re: Redesigning parallel dump/restore's wait-for-workers logic

This patch applies with a few minor offsets, compiles without
warning, and passes all regression tests including `make
check-world` with TAP tests enabled. This and the related patch
dealing with the parallel API definitely make things cleaner and
easier to follow.

I find it disappointing that ACLs continue to be held back as a
fourth step outside the framework of the "normal" ordering. That
is an ugly hack, which makes it impossible to, for example, create
a fifth step to create indexes on materialized views and refresh
them in anything resembling a clean fashion. Would it make sense
to deal with the ACL ordering hack in one of these patches, or
should that be left for later?

--
Kevin Grittner
EDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company