Using appsink with pad probes is not great.
Pull request: https://github.com/WebKit/WebKit/pull/587
This is quite risky actually. If we move to the GstVideoSink base class we loose the unlock/unlock_stop implementations of appsink. We would also loose some flushing handling that is handled in appsink already. I'd keep this improvement for later. There are too many tasks more important than this one. For now we can handle the video meta in the existing pad probe, without breaking anything in the critical video rendering code.