[FLINK-31753] Support DataStream CoGroup in stream mode with similar performance as DataSet CoGroup

[lindong28]

What is the purpose of the change

Add util methods that allow algorithm developers to co-group two DataStreams with the same semantics and similar performance as DataSet#coGroup(...)

Here are the results of running the benchmark specified in FLINK-31753's JIRA description:

• DataSet#coGroup takes 27.6 seconds.
• DataStreamUtils#coGroup takes 31.5 seconds.

The DataStream is roughly 12.3% slower than DataSet. The performance difference should be negligible for real-word applications whose co-group function is non-trivial.

Notes:

• The benchmark is run with Flink 1.18 snapshot.
• Most classes under the sort folder are copied from the corresponding classes in apache/flink. We can remove these classese from apache/flink-ml after making the corresponding classes in apache/flink public.

Brief change log

Added the static method DataStreamUtils#coGroup(...).

Does this pull request potentially affect one of the following parts:

• Dependencies (does it add or upgrade a dependency): no
• The public API, i.e., is any changed class annotated with @Public(Evolving): no

Documentation

• Does this pull request introduce a new feature? yes
• If yes, how is the feature documented? JavaDocs

[lindong28]

@zhipeng93 Can you help review this PR?

[zhipeng93]

Thanks for the PR :) I Left some comments below. Please take a look.

[zhipeng93]

Thanks for the update. LGTM.

[lindong28]

@zhipeng93 Thanks for the review.