[SPARK-11370][SQL] fix a bug in GroupedIterator and create unit test for it

[cloud-fan]

Before this PR, user has to consume the iterator of one group before process next group, or we will get into infinite loops.

[cloud-fan]

Looks like we only use keyOrdering to do equality check, why not just use ==? The currentGroup and currentRow are from the same input, they must be both unsafe or safe, and == for UnsafeRow is faster than keyOrdering.compare.

cc @marmbrus

[marmbrus]

This is the whole row, not just the key. This allows us to do the equality check on the key columns only (which might short circuit) instead of doing a full projection on each row to extract the key columns.

[cloud-fan]

Ah, sorry I missed it

[SparkQA]

Test build #44518 has finished for PR 9330 at commit a8cc6b5.

• This patch fails RAT tests.
• This patch merges cleanly.
• This patch adds no public classes.

[marmbrus]

I think these are the same, and I prefer the idiomatic version.

[SparkQA]

Test build #44529 has finished for PR 9330 at commit 9282e48.

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

[SparkQA]

Test build #44560 has finished for PR 9330 at commit 28f959c.

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

[marmbrus]

Thanks, merging to master.