Adding the shared heap write barrier caused regressions on some
benchmarks. Presumably this is because the compiler can't merge the
fast paths of the generational and shared heap write barrier.

This CL therefore introduces a CombinedHeapBarrier that manually
unifies the fast path for the marking, generational and shared heap
write barrier. This should make the barrier easier to optimize for
the compiler. In particular it should help to ensure that page flags
don't need to be loaded multiple times in a single full write barrier.

Bug: chromium:1326446, v8:11708
Change-Id: Iacd487f1263491cf4c05f25e004233a52b7c45a6
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3644964Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Cr-Commit-Position: refs/heads/main@{#80656}