[inductor] Remove ROCm xfail on test_cum{sum,prod}_zero_dim (#118558)

Fixes #118540, fixes #118541

Since the zero-dim case reduces to a pointwise operation, we don't fallback on
ROCm.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/118558
Approved by: https://github.com/malfet
diff --git a/test/inductor/test_torchinductor_codegen_dynamic_shapes.py b/test/inductor/test_torchinductor_codegen_dynamic_shapes.py
index af25284..33761b8 100644
--- a/test/inductor/test_torchinductor_codegen_dynamic_shapes.py
+++ b/test/inductor/test_torchinductor_codegen_dynamic_shapes.py
@@ -293,9 +293,7 @@
 if TEST_WITH_ROCM:
     test_failures.update(
         {
-            "test_cumprod_zero_dim_dynamic_shapes": TestFailure(("cpu", "cuda")),
             "test_cumsum_dynamic_shapes": TestFailure(("cpu", "cuda")),
-            "test_cumsum_zero_dim_dynamic_shapes": TestFailure(("cpu", "cuda")),
         }
     )