The logic for ensuring regular publishing in worker threads was broken
by growing the number of queues dynamically
(https://crrev.com/c/2467844). The first task(s) would assume a too
small number of worker threads, thus would publish to late (or never
before running out of units). This creates a large backlog of
to-be-published results when all threads eventually finish execution.

This CL fixes this by updating the per-task limit of results to process
before publishing. The updated value is read atomically using relaxed
memory ordering to ensure minimal impact on performance.

R=thibaudm@chromium.org

Bug: chromium:1138784, v8:11005
Change-Id: I2d00e50148e64db67a6b1a9f219ba60a1f4432ac
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2484365Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70646}