Some tests want to invalidate part of the VM state after an optimization
has consumed the old state but before the code is installed.

The existing mechanism for this is --block-concurrent-recompilation
and %UnblockConcurrentRecompilation(). The former suspends optimization
right after PrepareJob, before the background ExecuteJob phase. The
intrinsic can then be used to unblock it again.

This was good enough so far because the main "consume" work used to
happen on the main thread. With concurrent inlining this is no longer
true and we need something else.

This CL introduces three intrinsics:

%DisableOptimizationFinalization turns off automatic finalization of
background optimizations.

%FinalizeOptimization() can then be called at an appropriate time to
manually finalize (and thus install) the code and reenable automatic
finalization.

In case one wants to perform some action on the main thread after the
concurrent optimization has finished but before it is finalized, one can
do so with the help of %WaitForBackgroundOptimization() (see tests).

In a followup CL I'm removing the old mechanism since it now seems
redundant.

Bug: v8:12041, v8:7790
Change-Id: Ib7195789105922eb7e4bff86dc5bc11e96a4f97b
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3071400
Commit-Queue: Georg Neis <neis@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#76190}