Small omission in type_sanity.sql

Hi,

I was playing around with splitting up the tablespace test in regress so
that I could use the tablespaces it creates in another test and happened
to notice that the pg_class validity checks in type_sanity.sql are
incomplete.

It seems that 8b08f7d4820fd did not update the pg_class tests in
type_sanity to include partitioned indexes and tables.

patch attached.
I only changed these few lines in type_sanity to be more correct; I
didn't change anything else in regress to actually exercise them (e.g.
ensuring a partitioned table is around when running type_sanity). It
might be worth moving type_sanity down in the parallel schedule?

It does seem a bit hard to remember to update these tests in
type_sanity.sql when adding some new value for a pg_class field. I
wonder if there is a better way of testing this.

- Melanie