Implement IntersectVisitor#visit(IntsRef) whenever it makes sense

[iverase]

In #13149, the method Implement IntersectVisitor#visit(IntsRef) was introduced which is a nice. trick to avoid multiple virtual calls to IntersectVisitor#visit(int).

This was only implemented for PointRangeQuery but it is useful for all implementation that expects hitting that code path. Therefore in this PR we proposed to implement this method wherever it make sense.

[jpountz]

This makes sense to me, if the visit() call is inlined, then feeding the IntsRef into either an int[] (BufferAdder) or a FixedBitSet (BitSetAdder) can be auto-vectorized. Otherwise it can't.

This makes me wonder if we should instead introduce BulkAdder#add(IntsRef) and implement these visit(IntsRef) methods by calling adder.add(intsRef) directly, this would make auto-vectorizing less likely to get disabled by an apparently innocent change.

[iverase]

Thanks @jpountz. I added the method to the BulkAdder. The good thing is that for the buffer case we can use System.arraycopy which is even better.

[iverase]

Unrelated test error:

./gradlew test --tests TestFeatureField.testStoreTermVectors -Dtests.seed=BEC58524A0EDCFC6 -Dtests.locale=kw-Latn-GB -Dtests.timezone=Pacific/Kosrae -Dtests.asserts=true -Dtests.file.encoding=UTF-8

[iverase]

I made BulkAdder sealed so we are sure they will always just two implementations.

[iverase]

Actually we can make BulkAdder a sealed interface and the implementations are Java records.

[jpountz]

Unrelated test error

@jimczi This test is failing because we go through a slightly different code path on the first doc of a segment, which generates a slightly different exception message it seems.

[jpountz]

I opened a PR for the test failure: #14146