Fp16 re ranking and prefetch support

[Pulkitg64]

Description

We are adding FP16 support in Lucene. In #16383 we added the initial fp16 encoding and in #16473 we added scalar quantization support. This PR adds the re-ranking, approximate similarity, and prefetch support that already exist for the fp32 format.

[github-actions]

This PR has not had activity in the past 2 weeks, labeling it as stale. If the PR is waiting for review, notify the dev@lucene.apache.org list. Thank you for your contribution!

[goankur]

@Pulkitg64 heads up — #16705 changes KnnVectorValues#prefetch so that prefetch(int ord, int count) is the primitive and the int[] form coalesces consecutive ordinals over it. That touches OffHeapFloat16VectorValues, RescoreTopNQuery, FullPrecisionFloatVectorSimilarityValuesSource and TestOffHeapVectorValues, which this PR also changes.

Whichever lands second will need a small rebase. If #16705 goes first, the array-form prefetch
override this PR adds to OffHeapFloat16VectorValues collapses into the inherited coalescing
implementation, so it would be strictly less code — happy to do that rebase for you. Ordering the
other way works too, I would just forward prefetch(int ord, int count) from the fp16 wrapper.

[Pulkitg64]

Thanks @goankur for the heads up.

[msokolov]

I have some (not actionable) nits about naming; otherwise this looks like a straightforward extension of existing APIs, +1. I'll merge soon; we'll see if the prefetch changes get in first :)

[Pulkitg64]

Thanks Mike!