palignr has a false dependency, so use punpckhbw. Plus optimize for AVX,
which can have 1 less instruction.

The logic for i16x8.extmul_high_u is slightly complicated (SSE case)
to handle register aliasing. This allows us to have the most flexible
register allocation, and be able to optimize the most

Bug: v8:11468
Change-Id: I221b2d7a79009edb6c4060c136fc35ee7aff08fa
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2757224Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73415}