Have a new macro-assembler function Pshufb that does not require dst and
src to be equal, since the vpshufb can have 3 operands. On SEE, if dst
and src are not equal, emit a movapd(dst, src). This saves a move on
AVX.

Small cleanup to use kScratchDoubleReg2 instead of requesting
for a tmp (x64).

Bug: v8:9561
Change-Id: I131ad0456b272da857350762582cac1fb240ae40
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2513868
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70917}