Try using MPICH on macOS to avoid crashes

[LourensVeen]

The CI isn't working on macOS because we're getting an unknown MPI error. This switches to MPICH on macOS, see if that works...

[LourensVeen]

Seems like it gets through test_c_distributed_implementation but now it's hanging on test_fortran_implementation. We need that CI machine in Leiden I think...

[rieder]

MPI on GitHub/macOS has been a mixed bag, it seems unpredictable whether it will work or not. If nowadays MPICH is a better option, fine!

[rieder]

Seems like it gets through test_c_distributed_implementation but now it's hanging on test_fortran_implementation. We need that CI machine in Leiden I think...

Maybe we could buy a M4 Mac Mini, and run some UTM virtual machines on it? Should be able to test both Mac and Linux setups, and could also emulate x86-64 Linux (at least currently, with macOS still supporting Rosetta).

[LourensVeen]

OpenMPI is more actively maintained than MPICH, and it has a lot more features. If I were building an HPC cluster it's certainly what I would prefer. But we're not testing MPI here, we just need something that works, and if that's MPICH then I'm happy to use it for the CI.

We could get a mac, but the mac CI is actually passing 😄. And I guess it won't emulate a CUDA GPU, will it? Because that's another reason to have our own machine, so that we can test the GPU codes.