Bug in SQL grouping shortcut

Seen in Druid 24.0 and Druid 25.0

With `kttm_simple` loaded from the `Demo 1` demo query observe this behavior:

```
SELECT
count(*),
(case
when TIME_PARSE("timestamp", NULL, 'Asia/Kolkata')

Something is not right here. My understanding is that `GROUP BY 2` is just a shortcut.

Originally reported here: https://apachedruidworkspace.slack.com/archives/C0303FDCZEZ/p1672226327432399

Start with the Demo 1 query using the kttm_simple datasource and compare GROUP BY 2 with the expanded CASE expression shown in the report. Trace the SQL planning or validation path that handles ordinal grouping, then add a regression test demonstrating that both forms behave equivalently; done means the ordinal form runs without the grouping error.

Written by the indexing model from the issue text.