[FLINK-29043] Reduce overhead for non-feedback HeadOperator mails

[yunfengzhou-hub]

What is the purpose of the change

Improve the performance of Flink ML iteration mechanism by reducing the overhead on HeadOperator.

While #144 has reduced the overhead brought by HeadOperator.endInput(), it is discovered that the implementation has only reduced the overhead of Mails related to feedback edge. The other mails, like operator events, still need to wait until timeout before they are able to be processed, which is equivalent to Thread.sleep(timeout). Thus we need to improve the implementation in #144 to reduce the overhead for all mails the HeadOperator would receive.

The performance improvement brought by this PR on the unit tests of algorithms with iterations is as follows.

	Before this PR [1]	After this PR [2]
LinearSVCTest	21.195 s	5.111 s
LogisticRegressionTest	25.164 s	5.852 s
KMeansTest	31.86 s	12.23 s
LinearRegressionTest	32.526 s	7.736 s

[1] https://github.com/apache/flink-ml/runs/7966776640?check_suite_focus=true
[2] https://github.com/apache/flink-ml/runs/7968880728?check_suite_focus=true

Brief change log

• Replace the lock and condition used in HeadOperator.MailboxExecutorWithYieldTimeout with a timer to schedule the timeout event.

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)
• Does this pull request introduce a new feature? (no)
• If yes, how is the feature documented? (N/A)

[lindong28]

Thanks for the PR. LGTM.