Fix race condition in Thread::PostTaskImpl, potential for UAF

When a task is posted to a Thread using PostTaskImpl, the target thread
might process the task and destroy itself immediately (e.g., if the task
calls Quit() and the thread owner destroys the thread).

Previously, PostTaskImpl would release the mutex before calling
WakeUpSocketServer(). This allowed the target thread to run, process the
task, and potentially be destroyed before PostTaskImpl could access ss_
(socket server) in WakeUpSocketServer(), leading to a
Use-After-Free/Destruction.

This moves the WakeUpSocketServer() call inside the mutex lock. This
ensures that the task is not visible to the target thread (via Get())
until PostTaskImpl has finished its work, effectively extending the
critical section to cover the wake-up call.

This issue was caught on the msan bot:
https://chromium-swarm.appspot.com/task?id=76216a8fa02bd411&o=true&w=true

Bug: None
Change-Id: I15af36aab965c1c9180d0c982d23b200de1d3472
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/445120
Commit-Queue: Evan Shrubsole <eshr@webrtc.org>
Auto-Submit: Tomas Gunnarsson <tommi@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Evan Shrubsole <eshr@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#46791}