[SPARK-8794][SQL] Make PrunedScan work for Sample

[viirya]

JIRA: https://issues.apache.org/jira/browse/SPARK-8794

Currently PrunedScan works only when followed by project or filter operations. However, even if there is a Sample between these operations and PrunedScan, PrunedScan should work too.

[SparkQA]

Test build #36549 has finished for PR 7228 at commit df82785.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[chenghao-intel]

I am not sure if I understand correctly, probably a better place to prune the column or push down the filter is in the Optimzier.

[cloud-fan]

agree with @chenghao-intel , currently we push down project through Union, Aggregate, Join, Limit and Sort, and Sample should be included too.

[viirya]

@chenghao-intel @cloud-fan you are right. I will update this later. Thanks.

[SparkQA]

Test build #36579 has finished for PR 7228 at commit 7e4ba76.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[marmbrus]

I think it would be better to put this test in catalyst. For example, FilterPushdownSuite.

[SparkQA]

Test build #36640 has finished for PR 7228 at commit ede7cd8.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[marmbrus]

Thanks! Merged to master.