Updated bundle adjustment in stitching module: 1) it minimizes reprojection error now, 2) it minimizes error over focal, aspect, p.p.x, p.p.y parameters. Refactored and updated warpers.

switched to gpu::remap in opencv_stitching

added stream support to downsample, upsample, pyrUp and pyrDown

added gpu::buildWarpPlaneMaps and gpu::buildWarpCylindricalMaps functions, integrated into stitching

GPU: updated upsample, downsample functions, added pyrDown, pyrUp, added support of 16S filtering; put spherical warper on GPU (from opencv_stitching)

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

GPU: TargetArchs -> added FEATURE_SET prefix.

added assertion after all kernels calls

2) NVidia tests for GPU
3) FD sample that uses NVidia's interface.

minor code refactoring

added buffered version of gpu::integral function and updated performance test (it still works too slow)