Instead of completely finishing sweeping in the slow path of allocation,
just sweep all pages for the current PagedSpace. This will help
in making main thread allocation concurrent, since there is no need
anymore to lock the allocation mutexes of other PagedSpaces.

Bug: v8:10315
Change-Id: I1cf76d94fa7a22e726fc71f49c2d5669e4a0598c
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2292306
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#68822}