Handle possible deadlocks and hangs during PeerConnection teardown

Address two issues related to thread shutdown and teardown when the
target thread (network/worker) has been instructed to quit before the
PeerConnection destruction has completed:

1. In ScopedOperationsBatcher::Run(), if the target thread drops tasks
(e.g., because it is quitting), the loop would previously run infinitely
because task_idx did not advance. We now detect this by verifying that
task_idx has progressed after the BlockingCall, and abort if it has not.

2. In MethodCall::Marshal and ConstMethodCall::Marshal, if the target
thread drops the task because it is quitting, the calling thread would
previously hang forever because event_.Set() was never called. We now
use absl::Cleanup to ensure the event is set on task destruction.

Bug: b/515407473
Change-Id: Iea8ec32b6a9e9850983ed563db07b5c386ad834d
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/479140
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Commit-Queue: Tomas Gunnarsson <tommi@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#47930}