Unref the input frame in all cases. Previously it was only unreffed when the
frame was used for output.
When we produce an output frame, take an extra ref because the frame is now
freed elsewhere.
Don't wait for a keyframe but try to decode as soon as possible. In most cases,
the demuxer will send a keyframe first and things will work fine, in other
cases, like RTP, we try to decode ASAP to provide quick feedback. Some formats
also have no keyframes (h264 with intra refresh).
Keep track of the context veriables and renegotiate when something changes. We
need to use separate values for this because not all values can be copied into a
GstVideoInfo.
Set the output interlacing based on the libav context.
Use the demuxer and decoder framerate to construct an output framerate.
Use demuxer and decoder par to construct an output framerate.
