It would seem like the PR #11623 had chewed more
than it wanted to, non-fips build shouldn't really
be forced to use slower crypto/sha256 even for
presumed "non-performance" codepaths. In MinIO
there are really no "non-performance" codepaths.
This assumption seems to have had an adverse
effect in certain areas of CPU usage.

This PR ensures that we stick to sha256-simd
on all non-FIPS builds, our most common build
to ensure we get the best out of the CPU at
any given point in time.