[WIP][SPARK-22497][SQL] Project reuse

[wangyum]

What changes were proposed in this pull request?

The below SQL will scan table1 twice. This PR reuse the p1 and scan table1 once.

with p1 as (select * from table1 where key < 100), 
s1 as (SELECT key, count(*) FROM p1 group by key), 
s2 as (SELECT key, count(*) FROM p1 where key > -100 group by key) 
select s1.* from s1 join s2 on s1.key= s2.key

How was this patch tested?

unit tests

[SparkQA]

Test build #83744 has finished for PR 19727 at commit 1c458b8.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
• case class ReusedProjectExec(override val output: Seq[Attribute], child: ProjectExec)
• case class ReuseProject(conf: SQLConf) extends Rule[SparkPlan]

[viirya]

Simply reusing ProjectExec doesn't really reduce the scan. The duplication execution of CTE is a well known issue. I've addressed it before. But seems no solution to deal all possible cases yet.

[gatorsmile]

CTE reuse can cause the performance regression. It is hard to address without considering the costs.