Non-SIMD fallback

[Martin1994]

Summary

This patch implements a non-SIMD fallback on non-AVX2 machine.

Testing

Tested with Kirby & The Amazing Mirror's loading screen and prologue on:

• x86 AVX2 machine with AVX2 routine
• x86 AVX2 machine with non-SIMD routine
• aarch64 machine with non-SIMD routine

I also tuned the code a little bit to run both AVX and non-SIMD routines and made sure their output matched.

Known issues

• 8 bit routine is not covered.
• SIMD implementation with NEON would be hard because many of the intrinsics are not available. Also there is at most 128 bit SIMD on ARM platform rather than 256 bits.