ALTER EXTENSION SET SCHEMA versus dependent types

I happened to notice that the comment for AlterObjectNamespace_oid
claims that

* ... it doesn't have to deal with certain special cases
* such as not wanting to process array types --- those should never
* be direct members of an extension anyway.

This struck me as probably broken in the wake of e5bc9454e
(Explicitly list dependent types as extension members in pg_depend),
and sure enough a moment's worth of testing showed it is:

regression=# create schema s1;
CREATE SCHEMA
regression=# create extension cube with schema s1;
CREATE EXTENSION
regression=# create schema s2;
CREATE SCHEMA
regression=# alter extension cube set schema s2;
ERROR: cannot alter array type s1.cube[]
HINT: You can alter type s1.cube, which will alter the array type as well.

So we need to do something about that; and the fact that this escaped
testing shows that our coverage for ALTER EXTENSION SET SCHEMA is
pretty lame.

The attached patch fixes up the code and adds a new test to
the test_extensions module. The fix basically is to skip the
pg_depend entries for dependent types, assuming that they'll
get dealt with when we process their parent objects.

regards, tom lane