Introduce a vectorized soarDistance function

[iverase]

We are computing soar distance in a two step process. First we compute the distance between the vector and the centroid using a vectorized function. Then we compute the residual using another vectorized function. Finally we compute the soar distance.

Looking at the distance and residual functions, they are actually executing very similar operations, therefore combining them in the same function should lead to a nice speed up. This commit replaces the method #soarResidual with a method call #soarDistance. Microbenchmarks shows a nice speed up in this operation:

Mac:

Ignacio Vera Sequeiros
  8:27 AM
Benchmark                              (dims)   Mode  Cnt   Score   Error   Units
SoarDistanceBenchmark.soarDistanceNew      96  thrpt    5  58.639 ± 0.401  ops/ms
SoarDistanceBenchmark.soarDistanceNew     768  thrpt    5   9.806 ± 0.338  ops/ms
SoarDistanceBenchmark.soarDistanceNew    1024  thrpt    5   7.414 ± 0.273  ops/ms
SoarDistanceBenchmark.soarDistanceOld      96  thrpt    5  38.164 ± 1.216  ops/ms
SoarDistanceBenchmark.soarDistanceOld     768  thrpt    5   6.223 ± 0.113  ops/ms
SoarDistanceBenchmark.soarDistanceOld    1024  thrpt    5   4.790 ± 0.135  ops/ms

GCP:

Benchmark                              (dims)   Mode  Cnt   Score   Error   Units
SoarDistanceBenchmark.soarDistanceNew      96  thrpt    5  93.833 ± 0.975  ops/ms
SoarDistanceBenchmark.soarDistanceNew     768  thrpt    5  10.478 ± 3.752  ops/ms
SoarDistanceBenchmark.soarDistanceNew    1024  thrpt    5   8.644 ± 0.136  ops/ms
SoarDistanceBenchmark.soarDistanceOld      96  thrpt    5  52.466 ± 2.436  ops/ms
SoarDistanceBenchmark.soarDistanceOld     768  thrpt    5   8.336 ± 0.117  ops/ms
SoarDistanceBenchmark.soarDistanceOld    1024  thrpt    5   5.290 ± 0.071  ops/ms

[elasticsearchmachine]

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

[john-wagster]

nice! lgtm