[Misc] Follow up on the missing comment in PR#4989. (#5437)

Co-authored-by: Ubuntu <ubuntu@ip-172-31-28-63.ap-northeast-1.compute.internal>
Co-authored-by: Xin Yao <yaox12@outlook.com>

python/dgl/nn/pytorch/sparse_emb.py

@@ -168,11 +168,14 @@ class NodeEmbedding:  # NodeEmbedding
             Target device to put the collected embeddings.
         """
         if not self._comm:
-            # embeddings are stored on the CPU
+            # For embeddings stored on the CPU.
             emb = self._tensor[node_ids].to(device)
         else:
-            # embeddings are stored on the GPU
-            # the following method also covers self._world_size = 0 or 1
+            # For embeddings stored on the GPU.
+            # The following method is designed to perform communication
+            # across multiple GPUs and can handle situations where only one GPU
+            # is present gracefully, a.k.a. self._world_size == 1 or
+            # 0 (when th.distributed.is_initialized() is false).
             emb = nccl.sparse_all_to_all_pull(
                 node_ids, self._tensor, self._partition
             )