[+] Added 128-bit transpose

a big patch; use special proxy types (Input/OutputArray, Input/OutputArrayOfArrays) for passing in vectors, matrices etc.

added downsample function into gpu module, refactored it a little bit, added guard for CUDA related include in cascadeclassifier_nvidia_api.cpp

removed test whether z is negative in gpu::solvePnpRansac (there is no need in this for tests passing)

*fixed some warnings
*finished gpu test port to gtest framework

removed CMakeLists.txt from gpu/test folder to avoid precompiled header associated warnings under Ubuntu

removed catch from GPU stereo_bp test as --gtest_catch_exceptions flags exists (but err msg isn't printed in case when there is no GPU)