Remove codec thread from MediaCodecVideoEncoder.

webrtc/modules/video_coding/codecs/vp8/simulcast_encoder_adapter.cc

Index: webrtc/modules/video_coding/codecs/vp8/simulcast_encoder_adapter.cc
diff --git a/webrtc/modules/video_coding/codecs/vp8/simulcast_encoder_adapter.cc b/webrtc/modules/video_coding/codecs/vp8/simulcast_encoder_adapter.cc
index ae1151162ac19376c1e6a2620b6d2b082df050f0..a4d10671ba21e9f819645ebd54682ed90e239b9d 100644
--- a/webrtc/modules/video_coding/codecs/vp8/simulcast_encoder_adapter.cc
+++ b/webrtc/modules/video_coding/codecs/vp8/simulcast_encoder_adapter.cc
@@ -149,6 +149,7 @@ int SimulcastEncoderAdapter::Release() {
   while (!streaminfos_.empty()) {
     VideoEncoder* encoder = streaminfos_.back().encoder;
     EncodedImageCallback* callback = streaminfos_.back().callback;
+    encoder->Release();

[stefan-webrtc, 2017/02/21 12:29:56]

This was a bug, right? We should add a unittest to ensure it doesn't regress.

[sakal, 2017/02/21 12:53:50]

In practice not since all the encoders had an undocumented feature of releasing
themselves in the destructor.

This problem is caught by various instrumentation tests after this change.
Specifically, it hits the DCHECK in the destructor of MediaCodecVideoEncoder. I
believe it would be possible to make a more focused unit test if you find that
important.

[stefan-webrtc, 2017/02/21 13:13:08]

If it is hit by tests then I think it's good enough.

     factory_->Destroy(encoder);
     delete callback;
     streaminfos_.pop_back();