- mixer rendering implemention
=> frame based (was per sources batch based)

- add backward signaling to Observer/Obsevable classes
=> This help mixer to index sources for layout them.

- mutex'ed frame publish (VideoGenerator).

- sinks creation are now done at right places.
=> one per mixer (new), one per camera, one per stream reception.

- VideoRTPSession is fully responsible to handle video pipeline,
between RX/TS streams.
=> exhibit enterConference/exitConference to be aknowledged by upper layers.

- VideoSendThread is not longer a «thread», renamed as VideoSender.

- videoMixer_ is now a shared ptr in Conference objects.
=> getVideoMixer() return a rew shared_ptr also.

- Conference is now responsible to trig video conference pipeline

- std::this_thread::sleep_for() is not usable before GCC 4.1

Note: VideoMixer is not yet ready in this commit.

Fix bug #29818.
Note: non-conference mode has been tested and works.
Conference mode may works, but some issues have been detected.

Add also auto-start on local camera at any video capable call.
Fix VideoGenerator class.

Note: bliting function works only with YUV420 planar pixel format.

- fix also some Copyright dates.

- Fix VideoIOContext writable flag.
- Pass options to avformat_open_input() in video decoder.

Make video preview a standalone feature, currently start/stoppable from Preference UI.

Thanks to the new custom_io flag in libavformat's SDP demuxer, we can manage
our own UDP transports for RTP and RTCP. This allows us to comply with
RFC 4961.

If an older version of libavformat is detected, we fallback to sending and
receiving on different sockets.

Fixes segfault on avcodec_frame_get_defaults

If you try and join a NULL thread, you will segfault.

This way VideoSendThread will be able to implement it and we avoid the
callback hackery we had before.

Decompose big run() method into getFrame (which may be either from camera
or RTP) and renderFrame.

This fixes the behaviour of the camera not being free if starting and
stopping the video too quickly.