Added compilation with NPP to GPU module(FindNPP script). Tested only on win32, but it does not influence on OpenCV compilation if WITH_CUDA flag unset.

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.