Add some small arm neon optimizations

[markreidvfx]

Neon has some non-SSE2 features we can take advantage of. Here are a few trivial lut1d/lut3d optimizations and cleanups.

• use blendv, floor and fma(fused multiply add) intrinsics were possible 
• use hardware support for f16 conversions

On my Apple M2 Max

ocioperf --transform tests/data/files/clf/lut3d_preview_tier_test.clf

Before

Process the complete image (two buffers):  For 50 iterations, it took: [41.1767, 40.4428, 40.4575] ms

After

Process the complete image (two buffers):   For 50 iterations, it took: [36.5989, 35.5489, 35.5699] ms

[markreidvfx]

There appears to have been some issues with the github actions infrastructure when I submitted this pull request. All the actions did pass on my personal repo before I sent the pull request, I think re-running the actions will fix.

[doug-walker]

I re-started the Actions for you.