Performance improvement in annotation lookup

[testn]

• Annotation performs deep equals and hashCode. Wrapping it with AnnotationWrapper to use System.identityHashCode and == makes it significantly faster especially tests with a large number of data providers like the issues in Severe thread contention while running large test with parallel methods #772
• Add NULL_MARKER to mark when a method does not have the annotation. It will save the lookup in such case.

[juherr]

Do you have any tests or benchmarks to share?

[testn]

I ran a cutdown version of #772 by changing it to 6,000 data providers. It seems to reduce the test time by 6 seconds.

[juherr]

If the test only takes 6sec, I think you can add it :)

[testn]

Well... what it cuts down is the time it takes to initialize the tests which take much more than the test time itself!

[juherr]

Ok, but you can share it even if it won't be run in the suite each time.
I think it is important to have something to check the fix now and something to check perf regressions later.

[testn]

That's fine. It does not change the behavior. This is just to improve the caching behavior in JDK15AnnotationFinder without changing anything else.

[juherr]

Closing this PR as obsolete.

JDK15AnnotationFinder no longer uses Pair<Annotation, ?> directly as the annotation-cache key, so the AnnotationWrapper optimization proposed here no longer applies. The related #772 contention issue was separately fixed by #1305.

Thanks for the original investigation and proposal.