Handle special case of libuv for Windows (deepspeedai#7064)

More information on libuv in pytorch:
https://pytorch.org/tutorials/intermediate/TCPStore_libuv_backend.html
Issue tracking the prevalence of the error on Windows (unresolved at the
time of this PR): pytorch/pytorch#139990
LibUV github: https://github.com/libuv/libuv

Windows error:
```
  File "C:\hostedtoolcache\windows\Python\3.12.7\x64\Lib\site-packages\torch\distributed\rendezvous.py", line 189, in _create_c10d_store
    return TCPStore(
           ^^^^^^^^^
RuntimeError: use_libuv was requested but PyTorch was build without libuv support
```

use_libuv isn't well supported on Windows in pytorch <2.4, so we need to
guard around this case.

---------

Signed-off-by: Logan Adams <[email protected]>

deepspeed/comm/torch.py

@@ -145,11 +145,22 @@ def has_reduce_scatter_tensor(self):
 
     def init_process_group(self, backend, timeout, init_method, rank, world_size):
         if not torch.distributed.is_initialized():
-            torch.distributed.init_process_group(backend,
-                                                 timeout=timeout,
-                                                 init_method=init_method,
-                                                 rank=rank,
-                                                 world_size=world_size)
+            if not required_torch_version(min_version=2.4):
+                # Windows torch builds do not come with lib_uv by default.
+                # More information here: https://pytorch.org/tutorials/intermediate/TCPStore_libuv_backend.html
+                use_libuv = False if os.name == "nt" else True
+                torch.distributed.init_process_group(backend,
+                                                     timeout=timeout,
+                                                     init_method=init_method,
+                                                     rank=rank,
+                                                     world_size=world_size,
+                                                     use_libuv=use_libuv)
+            else:
+                torch.distributed.init_process_group(backend,
+                                                     timeout=timeout,
+                                                     init_method=init_method,
+                                                     rank=rank,
+                                                     world_size=world_size)
         self.using_mpi = torch.distributed.get_backend() == 'mpi'
 
     @disable_compiler_collective