Add windows CI

[digama0]

The tests fail with a segfault and BUG #1502 though, so I should probably procure a working windows system to test. I used tinycc on a windows VM without problems, so I'm not sure what's wrong with the CI version.

[digama0]

Oh boy, this is going to take a while to fix. The assumption that sizeof(long) == sizeof(void *) is pretty deeply ingrained in the code for pntrString and nmbrString.

[wlammen]

Yep I know. In the upcoming doxy description of metamath this is described as a bug. Affected is for example the structure nullPntrString. A partial solution is using a union { size_t, void* } Sizeof can be used with macros if I remember right, so the integer type whose size is just below the sizeof(void*) can be found with some macro tango.

For those interested: a pointer can have a so called segment prefix, so it is for example 48 bits wide. There is usually no integer type (long or long long) that has this weird bit width.

To be fair, segmented memory models are not found that often any more, instead the so called flat model is widely used now.

Maybe the config.h generated in the build process, has some useful information.

[digama0]

I'm working on a refactor that instead assumes sizeof(void*) == sizeof(size_t). I am not so worried about uncommon architectures for which this assumption fails (in which case we would have to use uintptr_t instead), but win64 is not an out of date or out of support target.