uncataloged tables are a vestigial husk

While working on some code today, I noticed that RELKIND_UNCATALOGED
appears to serve no useful purpose.  In the few places where we check
for it at all, we treat it in exactly the same way as
RELKIND_RELATION.  It seems that it's only purpose is to serve as a
placeholder inside each newly-created relcache entry until the real
relkind is filled in.  But this seems pretty silly, because
RelationBuildLocalRelation(), where the relcache entry is created, is
called in only one place, heap_create(), which already knows the
relkind.  So, essentially, right now, we're relying on the callers of
heap_create() to pass in a relkind and then, after heap_create()
returns, stick that same relkind into the relcache entry before
inserting the pg_class tuple.  The only place where that doesn't
happen is in the bootstrap code, which instead allows
RELKIND_UNCATALOGED to stick around in the relcache entry even though
we have RELKIND_RELATION in the pg_class tuple. But we don't actually
rely on that for anything, so it seems this is just an unnecessary
complication.

The attached patch cleans it up by removing RELKIND_UNCATALOGED and
teaching RelationBuildLocalRelation() to set the relkind itself.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company