Simplify USE_PAIR_ACTIVATIONS / SqrClippedReLU::propagate_pair()
STC https://tests.stockfishchess.org/tests/view/6a66bbd3c054e285ae028e0d
LLR: 3.19 (-2.94,2.94) <-1.75,0.25>
Total: 733184 W: 189246 L: 189739 D: 354199
Ptnml(0-2): 1721, 77704, 208246, 77189, 1732
Besides the cleanups, AVX512 now does the correct permutations at load time instead of using the AVX2 ones and then having to permute again at run time. AVXVNNI now also uses propagate_pair() (just like AVX2) which for some unknown reason it wasn't before. An argument could be made for keeping the AVX512 version of SqrClippedReLU::propagate() if we think it may be useful again in the future?
closes https://github.com/official-stockfish/Stockfish/pull/7021
No functional change