Optimize count(*) logic

[Novemser]

It is said from TiDB that certain optimization of push down columns could result in some performance improvement.

See pingcap/tidb#5447

[birdstorm]

keep in mind that count() acts differently in mysql and spark.

[Novemser]

@birdstorm Shall you describe the differences here in details? Thanks.

[birdstorm]

@Novemser if you use count() in spark, the result is 0. It is different from count(*).

so you must be careful if you want to apply their logic directly.

[birdstorm]

@Novemser It seems that TiKV has now supported setting key_only flag in coprocessor/scanner since pingcap/tikv#2580, maybe we can take advantage of that.