lucene_snapshot: Implement IVFVectorsReader::getOffHeapByteSize

[ChrisHegarty]

This issue has been filed to track the implementation of getOffHeapByteSize in the new IVFVectorsReader format. Both in the lucene_snapshot branch, and main too?

Trivially, the #128451 PR added info for raw vectors, but left a comment about completing the implementation. The raw vectors was added so as to get tests passing again in the lucene_snapshot branch.

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[benwtrent]

For this, I think we should:

• Return the centroids index size + the size of postings lists to match 5% of vectors. This is a large number of vectors, but I think over estimating is better than under.

Likely, you could even have lower memory than this estimate and still keep from degrading performance exponentially in many scenarios (especially as we have seen that exploring just 0.5% of vectors can work for some datasets...)