GitLab

The default JsonPrefStore task runner is used instead of injecting one.
Remaining use cases didn't require sharing the task runner except one
(to flush), but CommitPendingWrite()'s async reply API already provides
this functionality and we use it in this CL instead of exposing the
entire task runner (which is also more readable then implicitly
depending on the impl flushing right away on its task runner).

This is another take on hanxi's https://crrev.com/c/1153632
after debugging it locally I realized RundownTaskCounter needs to
observe the notification synchronously (as the WaitableEvent prevents
observing the reply). A nested Runloop is also not suitable (ref.
https://crbug.com/318527 and in code comments).

Note: JsonPrefStore's optional constructor params were flipped since the task
runner is now the most likely optional parameter.

This CL is a precursor to:
https://crrev.com/c/1148959.

R=sky@chromium.org

Bug: 848615, 318527
Cq-Include-Trybots: luci.chromium.try:ios-simulator-full-configs;luci.chromium.try:linux_mojo;master.tryserver.chromium.android:android_cronet_tester;master.tryserver.chromium.mac:ios-simulator-cronet
Change-Id: I24c03cb2e49ec667e592d7a78722cdaf0884af36
Reviewed-on: https://chromium-review.googlesource.com/1163628
Commit-Queue: Gabriel Charette <gab@chromium.org>
Reviewed-by: Scott Violet <sky@chromium.org>
Cr-Commit-Position: refs/heads/master@{#581343}