Solve the problem seen when using recent versions of CMake (such as 2.8.11): OpenCVConfig.cmake violates the policy CMP0004.

Symptoms:
CMake Error at CMakeLists.txt:10 (add_executable):
Target "my_app" links to item "C:/Program Files " which has leading
or trailing whitespace. This is now an error according to policy CMP0004.

This problem occurs when building on Windows when we depend on libraries located (for example) in "C:\Program Files (x86)..."

WITH_WINRT -> ENABLE_WINRT_MODE;
Some temporary char* replaced with Ptr<char>;
Build fix for regular WIN32;
Windows Platform SDK and MSVC search added to cmake;
Warinig fixes.

Additional CMAKE flag WITH_WINRT added.

Also, removed needless check of BUILD_SHARED_LIBS in core; in that case,
only the current module's library type matters.

3.0 seems like an appropriate time to do that.

* removed 2.0 PTX from default parameters since it has become obsolete
* added 3.5 BIN to prevent JIT compilation for super-Kepler

Now the HAVE_* macros are set in cvconfig.h. Most of them already were there,
but were nonfunctional.

Now the HAVE_* macros are set using cvconfig.h. Previously most
of them already were there, but were always undefined. One,
HAVE_COCOA, I had to add.

This also makes the CMake code more consistent; now, WITH_* variables
are always checked in cmake/*, while HAVE_* variables are checked in
modules/highgui/CMakeLists.txt.

I've left HAVE_LIBPTHREAD as a CMake variable, since TBB's build depends on it.

Some macros in internal.hpp depended on HAVE_ALLOCA_H, but they were, in turn,
unused, so I've just deleted them all.

I don't know why it didn't work for the original author, but it definitely
works now.

There is no WIN64 defined in the environment.

In these cases, just using print() is enough to ensure py2/3
compatibility.

OPENCLAMDBLAS and OPENCLAMDFFT never detected under linux.
lib64/import and lib32/import is the path on Windows but not Linux.
for CLAMDBLAS library we should use CLAMDBLAS_PATH (not CLAMDFFT_PATH)

This has no bearing on compilation, but it makes them show up in IDEs.

removed -fvisibility-inlines-hidden compiler option for CUDA sources

Also, removing explicit include path configuration,
since QT_USE_FILE takes care of that.

We don't really need it, it makes the code longer, and it can lead to
inconsistent paths when OpenCV is itself inside a symlink.

setting 'char' to be signed by default since some tests fail when it's wrong (e.g. native compilation on ARM Linux)