PATCH to allow concurrent VACUUMs to not lock each other out from cleaning old tuples

The attached patch allows VACUUMS's on small relations to clean up dead
tuples while VACUUM or ANALYSE is running for a long time on some big
table.

This is done by adding a "bool inVacuum" to PGPROC and then making use
of it in GetOldestXmin.

This patch is against current CVS head, but should also apply to 8.0.2
with minorpach warnings.

--
Hannu Krosing <hannu(at)tm(dot)ee>