[FEA] Conditional Channel Support

We're having issues with our DGL dependency since DGL uses conda labels to split CUDA version so we need a different channel specification per CUDA version, which `dependencies.yaml` currently does not allow.  By adding support, we would be able to better handle DGL testing for CUDA 12.

Related: https://github.com/dmlc/dgl/issues/7344 https://github.com/rapidsai/cugraph/pull/4346