Windows x64-to-x32 cross-build fails to match the stream_launched_host_function_adapter template

[eyalroz]

Splitting off a second issue mentioned in #295:

When cross-compiling on windows x6_64 for a x86 (32-bit) platform, MSVC shows the following error:

no instance of function template `cuda::stream_t::stream_launched_host_function_adapter matches the required type

I suspect this might have something to do with the following snippet in cuda_runtime_api.h:

#ifdef _WIN32
#define CUDART_CB __stdcall
#else
#define CUDART_CB
#endif

so perhaps I need to prefix the static method definition with CUDART_CB.

[eyalroz]

Fixed by the last commit.