[Native][x64] Native implementation of DiskBBQ scoring (distance w adjustment)

[ldematte]

In #140264 we introduce native functions to compute distance metrics for (Disk)BBQ. We kept the Panama implementation of the scoring, i.e. application of adjustments over the basic distance computation.
In this PR we introduce native functions that implement this second part too; this first iteration has optimized functions for x86 (AVX2 and AVX-512).
Benchmarks show an additional improvement of 25-30% (or, in other words, ~55% overall wrt the Panama implementation).

The PR also has a small refactoring on the C++ code, moving capabilities check to a separate file and some shared functionality to common headers.

[thecoop]

Are we able to use these methods for D2Q4 as well, or do those need completely separate implementations?

[tteofili]

Looks like this will need some rework after #141598, but I think the native code is already OK (it already gets the bulk size as a separate parameter and handles the tail). @tteofili can you confirm?

sorry for the late reply, I guess you figured the answer was yes :)

[tteofili]

LGTM !

[benwtrent]

Could we do a rename and call the actions that apply corrections applyCorrections?

[ldematte]

Are we able to use these methods for D2Q4 as well, or do those need completely separate implementations?

I added a index bits params for this reason - I'll double check, but the implementations looks to be identical

Or in other words, it is my intention to have 1 implementation for all

[thecoop]

Is it worth just doing the D2Q4 impl here as well then?

[ldematte]

Is it worth just doing the D2Q4 impl here as well then?

Yep, it was easy. Done!
I left D4Q4 alone as I know you are working on it; better if you address it then.

[ldematte]

Could we do a rename and call the actions that apply corrections applyCorrections?

Done!

[thecoop]

LGTM!