GraphBLAS: Support jitifyer without CMake for MinGW

[mmuetzel]

These modifications make the JIT compiler work without CMake for me using MinGW compilers (from MSYS2).
I don't know enough about MSVC. So, I left it at using CMake.

With those changes, I see the following timings for the ctests of LAGraph:

• first run: Total Test time (real) = 170.06 sec
• second run: Total Test time (real) = 17.82 sec

Compared to without the patch:

• first run: Total Test time (real) = 647.04 sec
• second run: Total Test time (real) = 19.04 sec

So, it significantly speeds up the first run. (The difference in the respective second runs is probably "random". Maybe caused by different load or alike.)

I believe this should still work on Linux or macOS without any noticeable difference. But I didn't actually try.

Some of these changes are pretty MSYS2-specific (e.g., fixing relocation issues). So, I'd understand if you prefer not to take it (or only take parts?). In that case, I'd propose a similar patch to MSYS2 that could be applied downstream when building GraphBLAS for distribution.
It would be easier though if MSYS2 wouldn't need to maintain a separate patch.

[DrTimothyAldenDavis]

Looks great! I'll merge it in and then give it a try on my various Linux systems and Macs (I have both an Intel Macbook and an M1 Macbook).

The LAGraph ctests can still pass even if there's a compiler error. If the kernel doesn't compile, I assume it's some sort of error in the user-defined type or operator. Or it might be a broken flag or something. If that happens, I can still go forward and do the work in the generic kernel instead, and I turn off the JIT so the compiler errors don't keep happening, as a fallback.

So I'll need to check it out carefully to make sure it doesn't break on Linux/Mac. It looks fine, but if something comes up I can always #ifdef it back to the original for Linux/Mac, and this your updates for MinGW / MSYS2.