[FEA] Remove extraneous modules/functionality in cudf namespace

[mroeschke]

Is your feature request related to a problem? Please describe.
When import cudf is called, we import a few modules/functions for setup but leave a lot of these objects in the namespace afterwards. Since we advertise cudf as a public namespace, we should only have objects that are designed to be use publicly

e.g. today you can do

In [5]: import cudf

In [6]: cudf.cupy.zeros(1)
Out[6]: array([0.])

pandas had a similar discussion in pandas-dev/pandas#30296

Describe the solution you'd like
Remove objects in the cudf namespace that are scoped to just cudf Python and are designed to be public

I suggest del-ing

• _setup_numba
• validate_setup
• numba_config
• cuda
• cupy
• rmm
• rmm_cupy_allocator
• RMMNumbaManager
• core
• clear_cache

And these objects which probably should not have public API use

• dtype
• BaseIndex
• isclose

Describe alternatives you've considered
Status quo.

Additional context
Add any other context, code examples, or references to existing implementations about the feature request here.