changed buffer type in linear filters to float.
added support of 1 channel image to linear filters.
added support of BORDER_REFLECT101, BORDER_REPLICATE and BORDER_CONSTANT border type to gpu linear filters.
minor fix in tests.
update comments in gpu.hpp.

module reorganization: added folder with pure device functions, cuda_shared.hpp renamed to internal_shared.hpp

added first version of gpu::matchTemplate, currently it works only with 8UC1 images and SQDIFF method

fixed compile error in transform.

added support of BORDER_REFLECT into gpu::cornerHarris and etc, added support of border extr. int linear filters

added first version of gpu::countNonZero for all data types, it doesn't support compute capability 1.0 yet, also fixed some little bugs