bug: incorrect batch query result when grouping by a non-continuous subset of table pk

[stdrc]

Describe the bug

No response

Error message/log

No response

To Reproduce

create table t (a int, b int, c int, primary key (a, b, c));
insert into t values (1, 1, 1), (1, 2, 3), (1, 3, 1);

select count(*) from t group by a, c; -- correct
select a, c, count(*) from t group by a, c; -- correct
select a, c, first_value(b order by b), count(*) from t group by a, c; -- correct
select a, c, first_value(b order by b), count(*) from t group by a, c having a = 1; -- incorrect!!!

Explain messages of the last two queries above:

 BatchExchange { order: [], dist: Single }
 └─BatchSortAgg { group_key: [t.a, t.c], aggs: [first_value(t.b order_by(t.b ASC)), count] }
   └─BatchExchange { order: [t.a ASC, t.c ASC], dist: HashShard(t.a, t.c) }
     └─BatchSort { order: [t.a ASC, t.c ASC] }
       └─BatchScan { table: t, columns: [a, b, c] }

 BatchSortAgg { group_key: [t.a, t.c], aggs: [first_value(t.b order_by(t.b ASC)), count] }
 └─BatchExchange { order: [t.a ASC, t.b ASC, t.c ASC], dist: Single }
   └─BatchScan { table: t, columns: [a, b, c], scan_ranges: [a = Int32(1)] }

It seems that the LogicalAgg is wrongly converted to BatchSortAgg when group by column is a non-continuous subset of table pk (input order), and when having clause is added, the optimizer generates a local batch plan without the BatchSort before Exchange.

Expected behavior

No response

How did you deploy RisingWave?

No response

The version of RisingWave

No response

Additional context

No response