Optimize:
- i32x4.widen_high_i16x8_s
- i32x4.widen_high_i16x8_u
- i16x8.widen_high_i8x16_s
- i16x8.widen_high_i8x16_u

These optimizations were suggested in http://b/175364869.

The main change is to move away from palignr, which has a dependency on
dst, and also the AVX version is 2 bytes longer than the punpckhqdq.

For the signed and unsigned variants, we have slightly different
optimizations. Unsigned variants can use an punpckh* instruction with a
zero-ed scratch register, that effectively zero-extends. Signed variants
use the movhlps instruction to move high half to low half of dst, then
use packed signed extension instructions.

The common fallback for these instructions is to use pshufd, which does
not have a dependency on dst, but is 1 byte longer than the punpckh*
instructions.

FIXED=b/175364869

Change-Id: If28da2aaa8f6e39a58e63b01cc9a81bbbb294606
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2591853Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71856}