bug: Two phase topn batch plan generation is incorrect when input is singleton.

[liurenjie1024]

Describe the bug

When generation 2 phase topn plan for batch query, and the input is singleton, the result is incorrect.

Error message/log

No response

To Reproduce

CREATE TABLE t1 (
    pk int, 
    a int, 
    b int,
    c bigint,
    d int
);

CREATE MATERIALIZED VIEW t1_mv AS
SELECT SUM(a) as sa, SUM(b) as sb, SUM(c) as sc, SUM(d) as sd FROM t1;

EXPLAIN
SELECT sa, count(*) as cnt2 
FROM t1_mv GROUP BY sa ORDER BY cnt2 DESC LIMIT 50 OFFSET 50;

The result is:

dev=> EXPLAIN
SELECT sa, count(*) as cnt2
FROM t1_mv GROUP BY sa ORDER BY cnt2 DESC LIMIT 50 OFFSET 50;
                                QUERY PLAN
--------------------------------------------------------------------------
 BatchExchange { order: [count DESC], dist: Single }
 └─BatchTopN { order: [count DESC], limit: 100, offset: 0 }
   └─BatchSimpleAgg { aggs: [internal_last_seen_value(t1_mv.sa), count] }
     └─BatchScan { table: t1_mv, columns: [sa] }
(4 rows)

Expected behavior

The topn should still be offset 50, limit 50.

How did you deploy RisingWave?

Docker.

The version of RisingWave

dev=> select version() ;
                                      version
------------------------------------------------------------------------------------
 PostgreSQL 9.5.0-RisingWave-1.7.0-alpha (fb42759622c52f682c6b905d887f8c2c3591c80d)
(1 row)

Additional context

No response

[liurenjie1024]

I think this bug is this line, we should return self rather input.

cc @chenzl25