Use NumPy functions: `normalize_axis_index`, `normalize_axis_tuple`

The functions `normalize_axis_index`, `normalize_axis_tuple` are not documented, but the author says
> When I added this function, it was always my intent for it to be consumed by downstream packages, but as Sebastian remarks, it wasn't really desirable to put it in the top-level namespace.
>
> I think I would be reasonably happy to make the guarantee that it would not be removed (or more likely, moved) without a lengthy deprecation cycle.
>
> Perhaps worth opening a github issue, so we can keep track of how many downstream projects are already using it.

http://numpy-discussion.10968.n7.nabble.com/Is-numpy-lib-shape-base-normalize-axis-index-considered-part-of-the-public-API-td48016.html#a48020

- SciPy uses `normalize_axis_index` but does not use `normalize_axis_tuple`.
- Dask uses `normalize_axis_index` and `normalize_axis_tuple`, while they are adapted directly from NumPy.
https://github.com/dask/dask/blob/2020.12.0/dask/array/numpy_compat.py

The scope of the issue is Python-level. CuPy should keep own Cython (`cdef`) implementations for performance, like NumPy has C implementations, too.

Start by comparing NumPy's normalize_axis_index and normalize_axis_tuple with the Python-level CuPy API, then review Dask's dask/array/numpy_compat.py usage. Done means the requested Python-level CuPy functions are available for downstream packages and documented, while the existing Cython implementations remain in place for performance.

Written by the indexing model from the issue text.