Support for MPS (Mac GPU) when using torch backend.

Hey Wessel!

I am just wondering if it might be possible to add support for the MPS backend in PyTorch as an alternative to Cuda, which would allow for the use of `lab` on the GPU on Mac. 

It is hopefully quite a simple change I think, I have tried an initial implementation #11 