size_limit option in copy

[youngsofun]

Summary

multi files load in parallel
row-based counting makes it harder.

[BohuTANG]

Can we change the parallel to serialized when the size_limit is setting?

[youngsofun]

maybe we can change a shard remain-rows each time a file/split/block is generated when parse in parallel. ( like push-down)
and add a resize pipeline to 1, to do the final size_limit.