Zero-copy access to VectorFloat<?> vectors backed by float[] arrays

[reta]

Problem

At the moment, the general patterns to extract float[] arrays from VectorFloat<?> instances are as such:

• call length() + get(index) to have a full copy, using per-element access
• call VectorFloat<?>::get() and check if it instance of float[] (covered by ArrayVectorFloat)
• call VectorFloat<?>::get() and check if it instance of MemorySegment , use reflection (since API is only available on JDK-22) to access heapBase() or ValueLayout (see please https://github.com/opensearch-project/opensearch-jvector/pull/507/changes#diff-af8a9e9960c5820cdc11161a01241bc490fb4aa17184c0e6b5397ae10af85307R18 as an example)

The reality is that all current implementations of VectorFloat<?> (but not necessarily future ones) are backed by float[] and could provide fast, zero-copy access to it.

Possible solutions

New API, opt-in only

Add FloatArray interface to allow easy float[] vector extraction from MemorySegmentVectorFloat and ArrayVectorFloat instances. It solves the problem by directly exposing the backed array through float[] array() method, no copying. This is opt-in only, the change is 100% non-breaking.

See please: #694

Change existing API

Add toArray method to existing VectorFloat<?> API. This is simpler change (since we could provide the default implementation using length() + get(index)) but we cannot guarantee that every toArray implementation is going to be zero copy, although we could document that.

See please: #697

CC @ashkrisk @akash-shankaran