Fix race between enabled() and set_enabled() in VideoTrack.

Along the way I introduced VideoSourceBaseGuarded, which is equivalent
to VideoSourceBase except that it applies thread checks. I found that
it's easy to use VideoSourceBase incorrectly and in fact there appear
to be tests that do this.

I made the source object const in VideoTrack, as it already was in
AudioTrack, and that allowed for making the GetSource() accessors
bypass the proxy thread hop and give the caller direct access.

Bug: webrtc:12773, b/188139639, webrtc:12780
Change-Id: I022175c4239a1306ef54059c131d81411d5124fe
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/219160
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Andrey Logvin <landrey@webrtc.org>
Commit-Queue: Tommi <tommi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#34096}

pc/video_track_source.cc

@@ -15,7 +15,7 @@
 namespace webrtc {
 
 VideoTrackSource::VideoTrackSource(bool remote)
-    : state_(kInitializing), remote_(remote) {
+    : state_(kLive), remote_(remote) {
   worker_thread_checker_.Detach();
 }