batch: singleton seq scan on local mode

[BugenZhao]

Describe the bug
Currently, when scanning from a singleton mview, we still schedule multiple scans.

To Reproduce

create table t (v int);
create materialized view mv as select v from t order by v limit 10;

explain verbose select * from mv;
                          QUERY PLAN                          
--------------------------------------------------------------
 BatchExchange { order: [], dist: [Single] }
   BatchScan { table: mv, columns: [mv.v], distribution: [] }

The distribution is [], that is, HashShard with empty keys. This is incorrect, we should use Single instead.

https://github.com/singularity-data/risingwave/blob/eb8109fc122ad51a8ccfcf4fc1efc03d79bfdb0a/src/frontend/src/optimizer/plan_node/batch_seq_scan.rs#L75-L93

#4153 introduced a fix while commented out, since it won't work on local mode now. We should fix this.

Expected behavior

                          QUERY PLAN                          
---------------------------------------------------------------
 BatchScan { table: mv, columns: [mv.v], distribution: Single }

Additional context
#4153 fixes the similar issue of StreamScan.
#3269 is related.

[BugenZhao]

@xxchan Could you please help to take a look?

[BowenXiao1999]

Well this problem is hard to find cuz the execution works fine and we can get the results. 👍

[BowenXiao1999]

Agreed that HashShard(vec![]) should not be a valid expression in our system.

[liurenjie1024]

This is a known issue, another case the table scan maybe singleton is partition prunning.

[liurenjie1024]

I think we should fix it in fragmenter, since we will move root fragment execution into frontend later.