fixed many bugs related to Mat::getUMat(), asynchronous kernel execution etc. Also, played a bit with ocl::cvtColor vs cv::cvtColor performance

In Python 3, version 3.3+, PyUnicode_AsUTF8() provides similar
functionality to Python 2's PyString_AsString().

In older versions of Python 3, there is no public function to provide
the same functionality.  However, the "internal" _PyUnicode_AsString()
does provide that functionality, so use it to replace
PyString_AsString().

With this patch, cv2 should compile for Python 3.[0-2].

removed refcount field from Mat; that helped to find and eliminate some memory leaks. perf tests should probably pass now

Conflicts:
	modules/core/include/opencv2/core/cuda.hpp
	modules/cudacodec/src/thread.cpp
	modules/cudacodec/src/thread.hpp
	modules/superres/perf/perf_superres.cpp
	modules/superres/src/btv_l1_cuda.cpp
	modules/superres/src/optical_flow.cpp
	modules/videostab/src/global_motion.cpp
	modules/videostab/src/inpainting.cpp
	samples/cpp/stitching_detailed.cpp
	samples/cpp/videostab.cpp
	samples/gpu/stereo_multi.cpp

Most importantly, added support for multiple include directores,
but also did some general cleanup.

Python 3 requires METH_VARARGS | METH_KEYWORDS, not just METH_KEYWORDS.

In these cases, just using print() is enough to ensure py2/3
compatibility.

Added a header with defines to enable the module to compile with either
python 2 or 3 without changes.

This protects from unwanted implicit conversions between types at compile time.

Also move cv::linemod to own header

Also, removed the one from modules/python/src2/cv.py and cleared its
executable bit, since it's not a script.