Fix out_tensor device in diag_test.py (#134020)

This benchmark fails if device='cuda' but out_tensor is on cpu

Pull Request resolved: https://github.com/pytorch/pytorch/pull/134020
Approved by: https://github.com/soulitzer

benchmarks/operator_benchmark/pt/diag_test.py

@@ -31,6 +31,7 @@ class DiagBenchmark(op_bench.TorchBenchmarkBase):
             "out": out,
             "out_tensor": torch.tensor(
                 (),
+                device=device,
             ),
         }
         self.set_module_name("diag")