sql: support column definition list with no data source alias

When a record-returning routine is used as a data source in a query, it must be given a column definition list to indicate the expected column types:
```
root@localhost:26257/defaultdb> create function f() returns record language sql as $$ select 1, 2 $$;
CREATE FUNCTION

Time: 44ms total (execution 24ms / network 21ms)

root@localhost:26257/defaultdb> select * from f();
ERROR: a column definition list is required for functions returning "record"
SQLSTATE: 42601
root@localhost:26257/defaultdb> select * from f() as foo (x int, y int);
x | y
----+----
1 | 2
(1 row)

Time: 3ms total (execution 2ms / network 0ms)
```
It is also valid to omit the alias for the data source, but we currently do not allow it:
```
root@localhost:26257/defaultdb> select * from f() as (x int, y int);
ERROR: a column definition list is required for functions returning "record"
SQLSTATE: 42601
```
This issue tracks adding support for a column definition list with no data-source alias.

Jira issue: CRDB-52293

Start by reproducing the two SQL examples in the issue and compare the aliased and alias-free forms. Done means a record-returning function accepts `as (x int, y int)` without a data-source alias and returns the declared columns and values, while the existing aliased form continues to work.

Written by the indexing model from the issue text.