[tensor-descriptor]: Enable test_make_tensor_descriptor_loop_carried (#4345)

Now that we have improved the translation of tensor descriptor (device)
APIs to block pointer APIs and fixed several functional issues in
downstream passes (e.g. materialze-block-ptr, remove-layout-conversions,
coalesceing, etc...) this PR enabled the test that checks the tensor
descriptor API in the presence of `scf.for`, `scf.while` loops
containing control flow (`scf.if`).

---------

Signed-off-by: Tiotto, Ettore <[email protected]>

Author: etiotto

python/test/unit/language/test_tensor_descriptor.py

@@ -666,8 +666,6 @@ def kernel_make_tensor_descriptor_loop_carried(a_ptr, M, N, MBLOCK: tl.constexpr
 @pytest.mark.interpreter
 @pytest.mark.skipif(is_hip(), reason="Currently unsupported by HIP devices")
 def test_make_tensor_descriptor_loop_carried(device):
-    if is_xpu():
-        pytest.skip("FIXME: issue #4132")
     M, N = 64, 512
     torch.manual_seed(42)
     A = torch.randn((M, N), dtype=torch.float32, device=device)