fix #18683 by enabling sort by aggregate

[tshauck]

Which issue does this PR close?

• Closes Schema error: No field named t.b. Valid fields are t.a. #18683

./target/debug/datafusion-cli 
DataFusion CLI v51.0.0
> create table t(a int, b int) as values (1,3), (3,4), (4,5);
0 row(s) fetched. 
Elapsed 0.021 seconds.

> select a from t group by a order by min(b);
+---+
| a |
+---+
| 1 |
| 3 |
| 4 |
+---+
3 row(s) fetched. 
Elapsed 0.017 seconds.

> EXPLAIN select a from t group by a order by min(b);
+---------------+-------------------------------+
| plan_type     | plan                          |
+---------------+-------------------------------+
| physical_plan | ┌───────────────────────────┐ |
|               | │       ProjectionExec      │ |
|               | │    --------------------   │ |
|               | │            a: a           │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │  SortPreservingMergeExec  │ |
|               | │    --------------------   │ |
|               | │  min(t.b) ASC NULLS LAST  │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │          SortExec         │ |
|               | │    --------------------   │ |
|               | │ min(t.b)@1 ASC NULLS LAST │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │       AggregateExec       │ |
|               | │    --------------------   │ |
|               | │       aggr: min(t.b)      │ |
|               | │        group_by: a        │ |
|               | │                           │ |
|               | │           mode:           │ |
|               | │      FinalPartitioned     │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │    CoalesceBatchesExec    │ |
|               | │    --------------------   │ |
|               | │     target_batch_size:    │ |
|               | │            8192           │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │      RepartitionExec      │ |
|               | │    --------------------   │ |
|               | │ partition_count(in->out): │ |
|               | │           1 -> 8          │ |
|               | │                           │ |
|               | │    partitioning_scheme:   │ |
|               | │       Hash([a@0], 8)      │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │       AggregateExec       │ |
|               | │    --------------------   │ |
|               | │       aggr: min(t.b)      │ |
|               | │        group_by: a        │ |
|               | │       mode: Partial       │ |
|               | └─────────────┬─────────────┘ |
|               | ┌─────────────┴─────────────┐ |
|               | │       DataSourceExec      │ |
|               | │    --------------------   │ |
|               | │         bytes: 224        │ |
|               | │       format: memory      │ |
|               | │          rows: 1          │ |
|               | └───────────────────────────┘ |
|               |                               |
+---------------+-------------------------------+
1 row(s) fetched. 
Elapsed 0.019 seconds.

Rationale for this change

This PR fixes a bug where if there's an aggregate expression in the order by that is not included in the projection, the query fails as the expression isn't propagated.

What changes are included in this PR?

Fixes the bug by making sure expressions in the order by are also passed to the aggregate function.

Are these changes tested?

Yes, tests are added to ensure the query compiles and produces the proper plan.

Are there any user-facing changes?

No

[dqkqd]

Should we add a check_columns_satisfy_exprs here for better UX?

            check_columns_satisfy_exprs(
                &order_by_post_aggr,
                std::slice::from_ref(&order_by_post_aggr),
                CheckColumnsSatisfyExprsPurpose::Aggregate(
                    CheckColumnsMustReferenceAggregatePurpose::OrderBy,
                ),
            )?;

Might need a testcase too.

> SELECT column2 FROM foo GROUP BY column2 ORDER BY column1
Error during planning: Column in ORDER BY must be in GROUP BY or an aggregate function