Instead of putting the 16 immediate bytes on the stack,
we move them into a temporary register.

The instruction-selector then has to change, to ensure that
the operands are distinct from the temporary.

Tested on the two workloads given in
https://github.com/zeux/wasm-simd/issues/2#issuecomment-614399004

For slow, the row "filter:" oct12 goes from ~50ms to ~27ms,
the rest of the figures look about the same or slightly faster.
For optimal, the same figure goes from ~25ms to ~24ms,
the rest of the figures look slightly faster.

Raw outputs are uploaded to bug.

Bug: v8:10117
Change-Id: I7f77a3066b5e24584f1c01574aa9311f56bd7fb4
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2152853
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Cr-Commit-Position: refs/heads/master@{#67190}