pg_depend take 2

Major changes since last revision:

initdb changes to account for table structure and new PIN elements for
genbki.sh. Enables really easy pinning of data in a table you know the
ID of.

Most objects enforced RESTRICT / CASCADE options.

Still needs code to track foreign keys between tables (right now it
catches the triggers -- but not the relation. See 'Foreign keys and
indexes' on hackers.

Serials sequences are dropped with the cascade.

'drop' regression test rewritten to test restrict / cascade of various
stuff. It may be missing a few things, but implicit cascades are
defined elsewhere (indexes, triggers, etc.)

Thoughts:
Do system objects depend on users? DROP USER ... CASCADE (drop all
tables, triggers, etc. that the user owns).

Will schemas auto-cascade to everything within them using the normal
DROP stuff or should all objects in a schema depend on the schema
implicitly? Allows RESTRICT / CASCADE of a schema fairly quickly.
(Tom?)