FFmpeg gets 1.372x faster RGB24-to-RGBA conversion with AVX-512

Think AI did this? Finding optimizations seems to be happening everywhere. Even llama.cpp is getting faster using AI, which they dont accept AI patches, so forks everywhere.