sql: improve type-checking of tuple comparison

In #94971 I'm introducing a regression logic test that results in an internal error when executed in a distributed fashion.

```sql
-- statement ok
CREATE TABLE t93396 (c1 TIME PRIMARY KEY, c2 INT8);
INSERT INTO t93396 VALUES ('0:0:0'::TIME, 0);

-- query error unsupported comparison: time to decimal
SELECT * FROM t93396 WHERE (c1, c2) = (SELECT 0.0, 0);
```
The problem is that we appear to not fully perform the type-checking of the tuple comparison on the gateway node, so when setting up a remote flow, this blows up.

Jira issue: CRDB-23250