Add CI on self-hosted runners

[epretti]

Run CI on self-hosted runners in addition to existing GitHub runners.

[epretti]

I have gotten this to a point where it is working.

There were a few issues I ran into. First, I had to install the CUDA 12 version of OpenMM since the CUDA 13 version causes pip to install a version of CUDA that is too new for the driver on the runner (which currently can't be upgraded further). OpenMM picks this one up instead of CUDA 13.0 installed in the image, which causes CUDA_ERROR_UNSUPPORTED_PTX_VERSION. Very strangely, this only breaks the FeNNix run and I don't know why. If there is a way to specify that CUDA 13.0 specifically should be installed by pip, that would be better.

Second, it was necessary to install an older version of GCC to build one of the packages in torchani's dependency tree. This is due to a bug in that dependency, which has been fixed, but torchani pins to an old version of a dependency, and the offending package doesn't have an ARM binary for the version with the bug, so it gets compiled from source.

[peastman]

Looks good. Hopefully we can upgrade the driver sometime so we can use CUDA 13.