Liftoff does not use all registers available on x64, so we can use
several hardcoded scratch registers instead of using the cache
registers which might need to be spilled.
This generates potentially smaller and more efficient code because we
need to spill and fill less.

R=titzer@chromium.org

Bug: v8:6600
Change-Id: I4ae20a1fb0ddd930d24130612825681752cfba24
Reviewed-on: https://chromium-review.googlesource.com/1146652Reviewed-by: Ben Titzer <titzer@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#54688}