[heap] Do not lock pending_allocation_mutex_ in CompactionSpaces

CompactionSpaces are only used during GC, so there is no need to
lock pending_allocation_mutex_ for them. Locking for GC allocations
actually caused multiple regressions.

Bug: chromium:1214765
Change-Id: I6db4ed96deced41dc52f04b2917ec944b4ccc674
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2928189Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74896}

src/heap/paged-spaces.cc

@@ -275,9 +275,10 @@ void PagedSpace::SetTopAndLimit(Address top, Address limit) {
          Page::FromAddress(top) == Page::FromAddress(limit - 1));
   BasicMemoryChunk::UpdateHighWaterMark(allocation_info_.top());
   allocation_info_.Reset(top, limit);
-  base::SharedMutexGuard<base::kExclusive> guard(
-      &heap_->pending_allocation_mutex_);
-  // The order of the following two stores is important.
+
+  base::Optional<base::SharedMutexGuard<base::kExclusive>> optional_guard;
+  if (!is_local_space())
+    optional_guard.emplace(&heap_->pending_allocation_mutex_);
   original_limit_ = limit;
   original_top_ = top;
 }