Fix ZeRO parameter initialization for tensors with `requires_grad=True` (#4138)

* Fix ZeRO parameter initialization for tensors with `requires_grad=True`

* Simplify detach logic

---------
Co-authored-by: NLogan Adams <114770087+loadams@users.noreply.github.com>

deepspeed/runtime/zero/partition_parameters.py

@@ -241,7 +241,7 @@ def zero_wrapper_for_fp_tensor_constructor(fn: Callable, target_fp_dtype: torch.
             kwargs['device'] = torch.device(get_accelerator().device_name(os.environ["LOCAL_RANK"]))
         tensor: Tensor = fn(*args, **kwargs)
         if tensor.is_floating_point():
-            tensor = tensor.to(target_fp_dtype)
+            tensor.data = tensor.data.to(target_fp_dtype)
 
         return tensor