Is your feature request related to a problem? Please describe.

AVX512 SIMD extensions should work even faster than AVX2 on modern architectures.
They should probably be only used on CPUs where thermal throttling is not an issue.

Describe the solution you'd like
New classifier implementations for AVX512.

Describe alternatives you've considered
It is worth considering whether this should always be enabled when available, or an approach similar to simdjson where older architectures have it disabled by default due to thermal throttling.