fixed possible video distortion problem when cap_avfoundation.mm is used to capture video (ticket #1980)

Added possibility in the class DetectionBasedTracker to work as a tracker only -- without the mainDetector. 
In this case new objects will be added into it by the method addObject(const cv::Rect& location).

which allows to pass as the second parameter of the method set any cv::Ptr<_Tp> object.

digits.py now saves the resulting SVM classifier for use in digits_video.py

added deskew for SVN and KNearest recognition sample