Can we optimize TaxonomyFacetSumIntAssociations#sumValues? [LUCENE-10373]

It looks like `TaxonomyFacetSumIntAssociations#sumValues` might also benefit from reduced `null` checks just like in #11386 (doing away with the use of the parent class `increment` method).

While we're at it, it would be nice to test moving to little endian encoding of the ord/value since we're trying to deprecate `BitUtil.VH_BE_INT`}.

The real challenge is that we don't have benchmark tasks for this class. Maybe we should add benchmarking first? I opened a `luceneutil` issue for this: https://github.com/mikemccand/luceneutil/issues/158

---
Migrated from [LUCENE-10373](https://issues.apache.org/jira/browse/LUCENE-10373) by Greg Miller (@gsmiller), updated Jan 18 2022
Attachments: [LUCENE-10373.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-10373/LUCENE-10373.patch)

Start at TaxonomyFacetSumIntAssociations#sumValues and compare the null-check optimization described in issue #11386, including the parent increment method. Review the linked luceneutil issue about benchmark tasks before evaluating little-endian ord/value encoding; done means measured benchmark results and a justified implementation decision.

Written by the indexing model from the issue text.