The destructor of the {WasmGCForegroundTask} can be called immediately
when scheduling that task (if the platform determines that the task can
never execute anyway). In that case, we deregister the task from the
wasm engine so we do not access it later (which would be UAF). This
deregistration leads to recursively taking a mutex now.
The only later access to the task happens to cancel the task. For this
purpose, we can also use the {CancelableTaskManager} of the isolate,
and avoid all code in the destructor. This should fix the reentrant
mutex, which leads to a DCHECK failure in debug builds and deadlock
in release builds.

R=mstarzinger@chromium.org

Bug: chromium:984970, v8:8217
Change-Id: I14f05a21ea961ecc391dc59af3b5eebf31e0f873
Cq-Include-Trybots: luci.v8.try:v8_linux_blink_rel
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1706480Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62804}