Fixed flann build with NEON;
Fixed Haming distance with NEON;
Honest cvRound for WinRT added;
cvRound test added;
Video IO with direct show disabled;

After setting depth to CV_32F for non double types this produces difference in results. I've attached test that reproduce this issue.The issue was found during anali broken GPU tests  for multiply/device functions.

Changed compiler flags for 32-bit MSVC;
Fixed handling of "d" suffix of debug binaries in run.py.

Optimized version of 32F Norm function demands bigger MaxErr value, because it calculates norm in floats instead of double.

added COVAR_ and SORT_ enums to core.hpp; fixed many, many VS2005, VS2010 and MinGW (GCC 4.5.2) warnings

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

reworked nearly all of the OpenCV tests (except for opencv_gpu tests) - they now use the Google Test engine.