Added initial implementation and test for the GPU version of add, based on NPP. Need to implement for other depths. Also waiting NPP at CMake.

added async version of postfilter_textureness and prefilter_xsobel, modified async test and added test for async version of stereobm 

* tests do not crash without test data
* test files renamed in systematic way
* added Test suffix for test classes names
* ts->printf used instead of some cout

meanShiftFiltering_GPU output parameters changed to CV_8UC4. This is a start for moving from 3 channel to C4 images within GPU module.

removed slamch.c/dlamch.c -specific GCC options, since the current implementation is very simple and relies on IEEE754 standard, rather than the actual machine behavior.

added flags OpenCV_COMPUTE_CAPABILITIES, which stores the version compute capability from which it is compiled opencv
now use opencv with cuda in your application  in the such way : set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${OpenCV_COMPUTE_CAPABILITIES})

renamed colorizeDisp to drawColorDisp, added acync version of drawColorDisp and reprojectImageTo3D_GPU.

Forgot to update so code after previous fix