cuda.core: fix _arr_is_c_contiguous discriminator for numba arrays (#1998)

`_arr_is_c_contiguous` checked `hasattr(arr, "flags")`, which is True
for both numpy arrays and numba `DeviceNDArray`. For numba `arr.flags`
is a plain dict, so the truthy branch falls into `arr.flags.c_contiguous`
and raises `AttributeError: 'dict' object has no attribute
'c_contiguous'`.

Discriminate on the flags object instead, mirroring the sibling
`_arr_is_writeable` helper. Unblocks six numba-cuda parametrizations:

- TestViewGPU::test_args_viewable_as_strided_memory_gpu[numba-cuda-{int8,float32}]
- TestViewGPU::test_strided_memory_view_cpu[numba-cuda-{int8,float32}]
- TestViewGPU::test_strided_memory_view_init[numba-cuda-{int8,float32}]

Made-with: Cursor

Author: Andy-Jost

cuda_core/tests/test_utils.py

@@ -102,7 +102,7 @@ def _arr_size(arr):
 def _arr_is_c_contiguous(arr):
     if torch is not None and isinstance(arr, torch.Tensor):
         return arr.is_contiguous()
-    return arr.flags.c_contiguous if hasattr(arr, "flags") else arr.flags["C_CONTIGUOUS"]
+    return arr.flags.c_contiguous if hasattr(arr.flags, "c_contiguous") else arr.flags["C_CONTIGUOUS"]
 
 
 def _arr_is_writeable(arr):