This is similar to the x64 optimization for i8x16.swizzle.

Use external refs to load the masks neded for i8x16.swizzle. Before it
would need 3 instructions (2 moves + 1 pshufd), now it requires 2 moves.
Also on AVX we can relax the dst == src requirement, which can
potentially save a move too.

Extract the code sequence into a macro-assembler function for sharing
between Liftoff and TurboFan.

Bug: v8:11346
Change-Id: Id0ec5e891595f0b0fc2922e932fc6c501eca8dc1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2727150Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73171}