[DiskBBQ] Speed up OSQ bit transformations

[benwtrent]

Description

Through benchmarking, I noticed that much of our time for quantizing query vectors (almost half?!?) we spend just transforming from flat int4 to the striped pattern of int4.

The code seems like it SHOULD be able to be vectorizable.

Or, maybe the solution is to transpose while getting the sum in sum-byte?

I am not sure, but its hilarious that spend almost as much time doing this loop as we do actually calculating the intervals and applying them to the vector.

[elasticsearchmachine]

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

[iverase]

The methods have been vectorized, closing.