- when IMREAD_UNCHANGED is set in flags, the tiff is read with 4 channels
- adding test for read/write

gen2.py reports headers it processes
cmake file restructured

The sanity check prevents Python 2 from calling findContours, because the
Python 2 bindings pass an empty matrix for `_contours`. If `_contours` is not
empty, the channel and depth check should determine whether the matrix is of
the correct type and size.

Note: the Python 2 bindings passes a matrix of type `STD_VECTOR_MAT` to
parameters of the type `OutputArrayOfArrays`.

Added an option to traincascade program to select the maximum number of threads to use during training

modified default stream initialization to allow concurrent calls modified cuda surf.cuda.cpp to allow concurrent call