[BUG] conda environment doesn't solve on arm64

[trxcllnt]

Attempting to solve the conda environment on arm64 machines fails with this error:

PackagesNotFoundError: The following packages are not available from current channels:
  - dgl

I don't see any aarch64 files in dglteam on anaconda.org, so I'm curious how the aarch64 conda packages are working? Should the dependencies.yaml pins be updated to match dgl >=2.4.0.th24.cu* used in the conda recipe?

Alternatively, we can restrict this dependency to x86_64:

--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -479,13 +479,13 @@ dependencies:
     specific:
       - output_types: [conda]
         matrices:
-          - matrix: {cuda: "12.*"}
+          - matrix: {arch: x86_64, cuda: "12.*"}
             packages:
               - "dglteam/label/th24_cu124::dgl"
-          - matrix: {cuda: "11.*"}
+          - matrix: {arch: x86_64, cuda: "11.*"}
             packages:
               - "dglteam/label/th24_cu118::dgl"
-          - {matrix: null, packages: ["dglteam/label/th24_cu124::dgl"]}
+          - {matrix: null, packages: null}
 
   depends_on_pyg:
     common:

[alexbarghi-nv]

Right now we can't really solve anything on arm64 because there are no appropriate packages for PyTorch.

[alexbarghi-nv]

If you need to run cugraph-pyg or wholegraph with arm64 we recommend the NGC containers.

[trxcllnt]

@alexbarghi-nv I don't necessarily need to run anything, I just need the environment to solve so I can compile the library on ARM. Right now I can create pip environments, but conda environments don't solve because this package isn't restricted to x86 in dependencies.yaml.

[alexbarghi-nv]

Ok, we need to restrict it to x86 then