[C10D] Document 'tag' limitation for nccl send/recv (#125278)

Existing documentation on isend/irecv also applies to send/recv. This PR
copies the doc/warning to send/recv ops as well.

Note: tag may be supplied, but will be ignored when used with nccl
backend.

Fixes #94819 #125079

Pull Request resolved: #125278
Approved by: https://github.com/kwen2501

torch/distributed/distributed_c10d.py

@@ -1878,6 +1878,9 @@ def send(tensor: torch.Tensor, dst: int, group: Optional[ProcessGroup] = None, t
     """
     Send a tensor synchronously.
 
+    .. warning::
+        ``tag`` is not supported with the NCCL backend.
+
     Args:
         tensor (Tensor): Tensor to send.
         dst (int): Destination rank on global process group (regardless of ``group`` argument).
@@ -1913,6 +1916,9 @@ def recv(tensor: torch.Tensor, src: Optional[int] = None, group: Optional[Proces
     """
     Receives a tensor synchronously.
 
+    .. warning::
+        ``tag`` is not supported with the NCCL backend.
+
     Args:
         tensor (Tensor): Tensor to fill with received data.
         src (int, optional): Source rank on global process group (regardless of ``group`` argument).