[+] Added 128-bit transpose

If a pipeline is specified manually, this has to be done in full now, ending with appsink. For example:

"uridecodebin uri=file:///path/to/some/video.avi ! ffmpegcolorspace ! appsink"

* removed unnecessary memcpy
* fixed minor memory leak

moved precomp.hpp before all other headers (to repair PCH mechanism). moved tables from the header to datamatrix.cpp

some more minor fixes in datamatrix: warnings eliminated, replaced MMX intrinsics (which are not recommended to use on x64) with SSE2.

Added to objdetect module: meanshift grouping (groupRectangles_meanshift(...) for input rectangles); new pedestrian detection model, trained on the Daimler base (getDaimlerPeopleDetector(...) ).
Some changes in the HOGDescriptor class interface(objdetect.hpp) (added useMeanShiftGrouping parameter to the detectMultiScale(...) function)