Fix compilation error on CUDA 13.0 and higher

[JonathanMaes]

This PR captures part of the changes introduced in mumax³-CO, as presented in C.-Y. You, J. Magn. 31, 204-213 (2026).

---

In CUDA 13.0, the number of arguments of cuCtxCreate(...) was changed, leading to a compilation error. Behind the scenes, 13.0 uses cuCtxCreate_v4 instead of the previously used cuCtxCreate_v2. Hence, this PR explicitly uses cuCtxCreate_v2 in context.go. However, this function is not normally exposed by cuda.h, so it is exposed manually in context.go via

//#include <cuda.h>
//CUresult cuCtxCreate_v2(CUcontext *pctx, unsigned int flags, CUdevice dev);
import "C"

Alternative

An alternative solution is

//#define __CUDA_API_VERSION_INTERNAL
//#include <cuda.h>
import "C"

since cuda.h (≥13.0) declares cuCtxCreate_v2 in a large if-block that only runs if __CUDA_API_VERSION_INTERNAL is defined.

I am not sure which would be the more robust option.

TODO

• Test both options on Linux