sql: pg_function_is_visible fails for UDFs

[knz]

Found while working on #88061.

Describe the problem

pg_function_is_visible() fails for the OID of UDFs.

To Reproduce

> CREATE FUNCTION myfunc(val INT) RETURNS INT CALLED ON NULL INPUT LANGUAGE SQL AS $$ SELECT val $$;
> select pg_function_is_visible(oid) from pg_proc where proname='myfunc';
ERROR: function 100105 does not exist: function undefined
SQLSTATE: 42883

Expected behavior

All the OID values in pg_proc are valid inputs to pg_function_is_visible.

cc @rafiss @mgartner @chengxiong-ruan for triage.

Jira issue: CRDB-23241
Epic: CRDB-23454

[rafiss]

i think this one will be fixed by #94771

[mgartner]

This looks like a regression introduced in acf5006. It works on v22.2.1:

defaultdb> select version();
                                        version
----------------------------------------------------------------------------------------
  CockroachDB CCL v22.2.1 (x86_64-apple-darwin19, built 2022/12/22 15:14:07, go1.19.1)
(1 row)


defaultdb> CREATE FUNCTION myfunc(val INT) RETURNS INT CALLED ON NULL INPUT LANGUAGE SQL AS $$ SELECT val $$;
CREATE FUNCTION

defaultdb> select pg_function_is_visible(oid) from pg_proc where proname='myfunc';
  pg_function_is_visible
--------------------------
            t
(1 row)

[rafiss]

I think the regression is in cee3ee9, not acf5006.

[mgartner]

Ahh, good catch!

[jordanlewis]

Yeah I happened to notice this as well, it's fixed here: 7221769

Will get this merged soon;

[rafiss]

also note that v22.2 has a different problem with pg_function_is_visible for UDFs. It's too permissive, since it doesn't actually inspect the search_path. See #89546

Backporting the fixes that we've made on master isn't feasible since it would require other changes, but I can make a smaller change just for v22.2.