perf: adds (some) arm NEON assembly for koalabear and babybear

[gbotrel]

Description

Follows #568 . Unfortunately Golang supports for ARM NEON instructions is limited and don't unlock satisfying results for Mul/ScalarMul/InnerProduct , need to revisit in the future or settle for sub-optimal implementation for these APIs if needed.

Sum/Add/Sub -->

Benchmarks

on c8g.4xlarge

VectorOps/sum_2097152-16            6.18ms ± 0%   0.19ms ± 0%  -96.90%  (p=0.333 n=2+2)
VectorOps/sum_4194304-16            12.4ms ± 0%    0.5ms ± 0%  -95.98%  (p=0.333 n=2+2)
VectorOps/sum_8388608-16            24.7ms ± 0%    1.1ms ± 0%  -95.42%  (p=0.333 n=2+2)
VectorOps/sum_16777216-16           49.4ms ± 0%    2.4ms ± 0%  -95.23%  (p=0.333 n=2+2)

VectorOps/add_2097152-16            1.53ms ± 0%   0.41ms ± 0%  -73.47%  (p=0.333 n=2+2)
VectorOps/add_4194304-16            3.08ms ± 0%   0.82ms ± 0%  -73.33%  (p=0.333 n=2+2)
VectorOps/add_8388608-16            6.11ms ± 0%   1.71ms ± 1%  -72.03%  (p=0.333 n=2+2)
VectorOps/add_16777216-16           12.2ms ± 0%    3.7ms ± 1%  -69.54%  (p=0.333 n=2+2)

VectorOps/sub_2097152-16            2.49ms ± 0%   0.41ms ± 0%  -83.62%  (p=0.333 n=2+2)
VectorOps/sub_4194304-16            4.98ms ± 0%   0.82ms ± 0%  -83.45%  (p=0.333 n=2+2)
VectorOps/sub_8388608-16            9.94ms ± 0%   1.71ms ± 0%  -82.80%  (p=0.333 n=2+2)
VectorOps/sub_16777216-16           19.9ms ± 0%    3.7ms ± 1%  -81.58%  (p=0.333 n=2+2)