* read "SeekPreRoll" from the source track (if available)
* use the longest track duration as segment duration, instead of the video track duration
* do not hardcode the "Cue" reserved space behavior
* do not hardcode the "EBML Void" element, unreported issue. The size was not properly calculated
* rewrite the key-frame picking
* remove writeInt(), writeFloat() and writeShort() methods, use inline code
* set "SeekPreRoll" and "CodecDelays" values on output tracks (if available)
* rewrite the "Cluster" maker
* rewrite the code of how "Cluster" sizes are written
Fix encode() method (the reason of this commit/pull-request):
* Use the unsigned shift operator instead of dividing the value, due precession lost