sql: track UPDATE SET column dependencies in UDFs and SPs

Related to #102771.

We don't currently track UDF and SP dependencies on `SET` column in `UPDATE` statements in the routine body. This allows these columns to be renamed or dropped, which breaks the routine. For example:

```sql
CREATE TABLE t (
a INT PRIMARY KEY,
b INT
);

CREATE FUNCTION up() RETURNS VOID LANGUAGE SQL AS $$
UPDATE t SET b = 11 WHERE a = 1
$$;

-- This should fail, but it succeeds.
ALTER TABLE t RENAME COLUMN b TO b2;

-- Now, this fails.
SELECT up();
-- ERROR: column "b" does not exist
-- SQLSTATE: 42703
```

Jira issue: CRDB-36895