[X86][AVX] Prefer vinsertf128 to vperm2f128 on AVX1 targets

Splatting the lower xmm with vinsertf128 is at least as quick as vperm2f128, and a lot faster on some AMD targets.

First step towards PR50053