DiversifyingChildren speedup

[alessandrobenedetti]

Description

The change is simple, arrays rather than an object + tests .

Some basic benchmarks (still to be fully reviewed):
./gradlew -p lucene/join test --tests "TestDiversifyingNearestChildrenKnnCollectorPerformancetestThroughput*" -Ptests.verbose=true -Ptests.filter=@nightly 2>&1 | grep -E "collect/sec|tests,|SUCCESS|FAIL" | head -15

Original

1> medium-k k= 32 children= 100000 avg= 2,47 ms 40,53 M collect/sec
1> high-k k= 100 children= 100000 avg= 2,57 ms 38,95 M collect/sec
1> small-k k= 10 children= 100000 avg= 3,29 ms 30,43 M collect/sec
1> dense-updates high k k= 100 children= 100000 avg= 3,56 ms 28,09 M collect/sec
1> dense-updates k= 10 children= 100000 avg= 2,90 ms 34,51 M collect/sec

PULL REQUEST

1> small-k k= 10 children= 100000 avg= 1,39 ms 72,17 M collect/sec
1> dense-updates k= 10 children= 100000 avg= 1,78 ms 56,33 M collect/sec
1> high-k k= 100 children= 100000 avg= 1,72 ms 57,99 M collect/sec
1> medium-k k= 32 children= 100000 avg= 1,48 ms 67,77 M collect/sec
1> dense-updates high k k= 100 children= 100000 avg= 2,74 ms 36,52 M collect/sec

[benwtrent]

Why isn't this a JMH benchmark? I am not sure about having "benchmark type things" in test files. JMH provides many of the things we need. (see: benchmark-jmh

[alessandrobenedetti]

I'm planning to remove these before the merge, just leaving them there now, in case you want to validate the benchmarks!

[benwtrent]

OK, sweet! I moved them to a JMH benchmark and validated. Yep, I am seeing 50-100% throughput improvements with this change on collecting :D.