Remove MPI from multi-GPU example

[jwallwork23]

Closes #253.

I decided to close #258 and split it into two separate PRs. Here's the first one that removes MPI from the multi-GPU example. (The second one will introduce a CPU-only example that uses MPI.)

Note that I decided to switch the names of the multigpu.py module / MultiGPU class back to simplenet.py / SimpleNet in this example. This is because the class is a direct copy. (Unlike in MultiIONet, where it's modified.)

[jwallwork23]

Tested on a laptop with 1 CUDA GPU device with the patch

diff --git a/examples/3_MultiGPU/multigpu_infer_fortran.f90 b/examples/3_MultiGPU/multigpu_infer_fortran.f90
index 297844e..cfba096 100644
--- a/examples/3_MultiGPU/multigpu_infer_fortran.f90
+++ b/examples/3_MultiGPU/multigpu_infer_fortran.f90
@@ -27,7 +27,7 @@ program inference
    type(torch_tensor), dimension(1) :: out_tensors

    ! Variables for multi-GPU setup
-   integer, parameter :: num_devices = 2
+   integer, parameter :: num_devices = 1
    integer :: device_index, i

    ! Get TorchScript model file as a command line argument
diff --git a/examples/3_MultiGPU/multigpu_infer_python.py b/examples/3_MultiGPU/multigpu_infer_python.py
index 1b49398..c504063 100644
--- a/examples/3_MultiGPU/multigpu_infer_python.py
+++ b/examples/3_MultiGPU/multigpu_infer_python.py
@@ -53,7 +53,7 @@ def deploy(saved_model: str, device: str, batch_size: int = 1) -> torch.Tensor:
 if __name__ == "__main__":
     saved_model_file = "saved_multigpu_model_cuda.pt"

-    for device_index in range(2):
+    for device_index in range(1):
         device_to_run = f"cuda:{device_index}"

         batch_size_to_run = 1