[BEAM-4783] Fix issues created in #6181.

[kyle-winkelman]

Fix issues created in #6181 and incorrectly fixed in #6884 (although this PR increased readability greatly).

Before any work on BEAM-4783, GroupCombineFunctions would always use a new HashPartitioner(rdd.rdd().sparkContext().defaultParallelism());.

The intent was to skip the creation of this Partitioner and call groupByKey() with no arguments only when the new bundleSize option was in use. #6181 actually did the opposite causing performance impacts, and because #6181 had terrible readability #6884 did not fix it correctly.

I am also hopeful that this can be cherry picked into 2.10 if another RC is created to get the Spark Runner's performance back to the levels seen in 2.7.

@iemejia @timrobertson100

---

Follow this checklist to help us incorporate your contribution quickly and easily:

• Format the pull request title like [BEAM-XXX] Fixes bug in ApproximateQuantiles, where you replace BEAM-XXX with the appropriate JIRA issue, if applicable. This will automatically link the pull request to the issue.
• If this contribution is large, please file an Apache Individual Contributor License Agreement.

It will help us expedite review of your Pull Request if you tag someone (e.g. @username) to look at it.

Post-Commit Tests Status (on master branch)

Lang	SDK	Apex	Dataflow	Flink	Gearpump	Samza	Spark
Go		---	---	---	---	---	---
Java
Python		---			---	---	---

[iemejia]

Thanks @kyle-winkelman sorry for not responding before to your messages. I was aware of the issue, just too busy with other stuff to take a look again. I will do and ping you when done.

[iemejia]

LGTM sorry for the delay, and thanks for bringing back this fix @kyle-winkelman.
I am not sure there is going to be a new RC but if there is I will try to include this one. Anyway next version (2.11.0) is going to be cut next week so it will be included there for sure.

[iemejia]

Merged manually to correct the commit title and add an extra @Nullable annotation.