Add bulk off-heap scoring for vector formats

### Description

We found some pretty awesome latency improvements with float32 vectors bulk scoring off-heap. It makes me think we can get nice improvements for byte vectors and all the quantized formats.

See: https://github.com/apache/lucene/pull/14980

&: https://github.com/apache/lucene/pull/14978

Start by reviewing the linked PRs 14980 and 14978 to understand the existing float32 off-heap bulk-scoring work and its entry points. Trace how that approach applies to byte vectors and the quantized formats, then verify that those formats support bulk off-heap scoring with the expected latency improvements.

Written by the indexing model from the issue text.