sql: allow * expressions in user-defined function (UDF) bodies

[mgartner]

In v22.2 we disallow * expressions in UDF bodies, see #86070. We should allow these expressions. We'll need to figure out how to handle cases where a table's schema changes after a function has been created.

Jira issue: CRDB-20579

Epic CRDB-19496

[mgartner]

In a previous issue #86070 (comment), @ajwerner said:

I think we can disable support for the * for this release and it would be okay. Postgres in views and in the eagerly resolved form of UDFs turns the * into column names at creation time. I think we should do that.

Here's an example of this behavior in Postgres (using the sql_body BEGIN ATOMIC syntax - we want this same behavior with the AS '<func_body>' syntax):

marcus=# CREATE TABLE t (a INT);
CREATE TABLE

marcus=# CREATE FUNCTION f() RETURNS INT LANGUAGE SQL BEGIN ATOMIC SELECT * FROM t; END;
CREATE FUNCTION

marcus=# INSERT INTO t VALUES (1);
INSERT 0 1

-- f() Returns the value of column a.
marcus=# SELECT f();
 f
---
 1
(1 row)

marcus=# ALTER TABLE t ADD COLUMN b INT DEFAULT 2;
ALTER TABLE

-- f() Still returns the value of column a.
marcus=# SELECT f();
 f
---
 1
(1 row)

marcus=# ALTER TABLE t RENAME COLUMN a TO c;
ALTER TABLE


-- f() Still returns the value of column c, which was
-- renamed from a.
marcus=# SELECT f();
 f
---
 1
(1 row)

marcus=# ALTER TABLE t RENAME COLUMN b TO a;
ALTER TABLE

-- f() Still returns the value of column c, even though
-- there is another column b.
marcus=# SELECT f();
 f
---
 1
(1 row)

[rharding6373]

At this point we've added support for basic * expansion, but do not have support for the schema change edge cases (e.g., column and table renaming) mentioned in #90080 (comment). That should be supported when #83233 is resolved.