Problem with using python MParT with GPU when GPU is being used

Someone else was using a GPU as well and so I got the error

>>> import mpart as mt
RuntimeError: cudaDeviceSynchronize() error( cudaErrorDevicesUnavailable): CUDA-capable device(s) is/are busy or unavailable /home/dannys4/git-repos/kokkos/core/src/Cuda/Kokkos_Cuda_Instance.cpp:151

despite there being multiple GPUs (one being unused). I had no luck trying to change the device via os.environ['DEVICE_ID'] = '1'. I will try changing the arguments within the startup file but, regardless, we should

1. Create a way so that MParT.py isn't initialized automatically (same for other bindings)
2. Document how to initialize it manually
3. Document what you can use to change the initialization
4. See why we aren't allowed to run Kokkos on GPU while something else is running

Start with the automatic initialization of MParT.py and the other bindings, then inspect the referenced Kokkos_Cuda_Instance.cpp path and the startup-file arguments. Reproduce the failure with multiple GPUs in use, determine how device selection and concurrent Kokkos use are configured, and document manual initialization and supported options. Done means initialization is controllable and the GPU concurrency behavior is explained.