[BUG] Compute memory_usage for list<struct<list>>

[GregoryKimball]

Describe the bug
I would like to use the df.memory_usage API with nested types in cuDF-python. The API works great with deeply nested lists and structs, but with mixed nesting as in list<struct<list>>, the API throws an exception.

Steps/Code to reproduce bug

# create column typed as "list<struct<'b':list<int>>>"
df = cudf.DataFrame({
    'a': [[{'b':[1]}]]
})
print(df.memory_usage(deep=True).sum())

Traceback (most recent call last):
...
  File "/opt/conda/envs/all_cuda-120_arch-x86_64/lib/python3.10/site-packages/cudf/core/dtypes.py", line 665, in itemsize
    return sum(
  File "/opt/conda/envs/all_cuda-120_arch-x86_64/lib/python3.10/site-packages/cudf/core/dtypes.py", line 666, in <genexpr>
    cudf.utils.dtypes.cudf_dtype_from_pa_type(field.type).itemsize
AttributeError: 'ListDtype' object has no attribute 'itemsize'

Expected behavior
I would expect the memory_usage function called with deep=True to work on any cuDF type.

Environment details
Docker container based on rapidsai/ci:cuda12.0.1-ubuntu22.04-py3.10, running on A100 DGX, built from source on commit e1f6c7cb3fa62 (~23.10)