Use 8x SIMD conversions when appropriate

[starkat99]

Use the 8x x86 SIMD conversions when appropriate, not just the 4x. These are _mm256_cvtph_ps and _mm256_cvtps_ph. Needs some profiling to determine which to use over the other and in what circumstances, possible that it might be better to use 8x always.

[Shnatsel]

The commit that closed this also sped up the fallback codepath, presumably because it's a loop instead of 4 independent calls to the fallback function, so there's less code bloat.

When will a new release with this change ship? We'd like to use it in the exr crate: johannesvollmer/exrs#191

[starkat99]

Was going to wait until doing some more ARM work, but went ahead and did a 2.2.1 release just now for you.