[BUG] Numba 0.57 will break row UDFs

[gmarkall]

Describe the bug

Identified by @brandon-b-miller:

numba/numba#8120 added a bitwidth attribute to the Record type, which is computed by converting the fields to their NumPy dtype and getting the item size. However, cuDF uses Record instances that have non-NumPy members - I believe we generate these types here: https://github.com/rapidsai/cudf/blob/branch-23.04/python/cudf/cudf/core/udf/row_function.py#L30

The consequence is that Numba raises when attempting to convert these members to NumPy dtypes.

Steps/Code to reproduce bug

Running the test_string_udf_strip test should reproduce the issue when Numba built from its main branch is in the environment should cause this: i.e. pytest test_udf_masked_ops.py -k test_string_udf_strip

Expected behavior

Successful execution of the test.

Environment overview (please complete the following information)

Any environment in which Numba built from current main is present (or the 0.57 RC / release when it's out).

Environment details

Not supplied.

Additional context

None.

[vyasr]

Resolved in #13271