Bitset iteration optimization and improve safety

[leventov]

• Deduplicate looking for bitset.nextSetBit() in BitSetIterator.next() and hasNext()
• Along the way it makes this iterator to follow Iterator contract, should throw NoSuchElementException on calling next() after hasNext() = false. Before it was returning -1, that is bug-prone.
• Add BitmapIterationTest which verifies Bitmap iteration implementations. BTW it fails for Roaring right now, but immediate fix is not possible, because it's a library dependency.

It makes iteration over a bitset 20-30% faster (before-after, benchmarks from #3751):

Benchmark                      (bitmapAlgo)  (prob)   (size)  Mode  Cnt        Score        Error  Units
BitmapIterationBenchmark.iter        bitset     0.0  1000000  avgt    5        4.052 ±      0.137  ns/op
BitmapIterationBenchmark.iter        bitset   0.001  1000000  avgt    5    36647.927 ±   1738.782  ns/op
BitmapIterationBenchmark.iter        bitset     0.1  1000000  avgt    5   699019.477 ± 158638.803  ns/op
BitmapIterationBenchmark.iter        bitset     0.5  1000000  avgt    5  3395422.304 ± 158315.241  ns/op
BitmapIterationBenchmark.iter        bitset    0.99  1000000  avgt    5  5875929.154 ± 241268.439  ns/op
BitmapIterationBenchmark.iter        bitset     1.0  1000000  avgt    5  5626511.995 ± 759176.569  ns/op

Benchmark                      (bitmapAlgo)  (prob)   (size)  Mode  Cnt        Score        Error  Units
BitmapIterationBenchmark.iter        bitset     0.0  1000000  avgt    5        3.221 ±      0.362  ns/op
BitmapIterationBenchmark.iter        bitset   0.001  1000000  avgt    5    26713.522 ±   6375.378  ns/op
BitmapIterationBenchmark.iter        bitset     0.1  1000000  avgt    5   521381.301 ±  16891.222  ns/op
BitmapIterationBenchmark.iter        bitset     0.5  1000000  avgt    5  2053421.442 ±  34196.751  ns/op
BitmapIterationBenchmark.iter        bitset    0.99  1000000  avgt    5  3969497.718 ± 410796.782  ns/op
BitmapIterationBenchmark.iter        bitset     1.0  1000000  avgt    5  4040926.926 ± 139893.245  ns/op

This is a copy of metamx/bytebuffer-collections#37

[leventov]

See RoaringBitmap/RoaringBitmap#129

[gianm]

👍 on the main code change but please adjust the test code as mentioned.

[gianm]

What exactly fails here? Is it bad that it fails? If it's not bad, why are we testing it?

Please add a comment here answering those questions (and linking anything we're waiting for upstream in Roaring) or else remove the tests if they're not testing something we actually care about.

[leventov]

@gianm done

[gianm]

thanks, LGTM

[fjy]

👍