[FEA] Phase out CFFI from libcudf

[jrhemstad]

Is your feature request related to a problem? Please describe.
Currently both CFFI and Cython bindings are being built for the libcudf API. CFFI is problematic because it requires us to enforce a pure C interface in the external libcudf APIs. Furthermore, we cannot have any include statements in the files parsed by CFFI.

Describe the solution you'd like
We should fully phase out CFFI in favor of Cython. In this way, new APIs can be made to be C++ and old APIs can be converted (if desired) from their C interfaces to C++ interfaces.

Additional context
As I understand it from @mt-jones, all of the libcudf bindings to cuDF are already updated to Cython, therefore we should have no problem fully removing CFFI.

[dantegd]

Responding with the status for full migration:

• Most bindings have been written/updated to Cython already, some (quite a few actually) I haven't pushed, my attention has been big time on the other side of the river lately (aka cuML). But this should be done not much after the holidays.
• We can already start adding C++ functionality in separate .hpp headers as suggested by a very smart person (aka yourself @jrhemstad ;) ). Then they can be wrapped with cython already skipping cffi.
• Full transition will happen before 0.5 is complete, but I would strongly encourage not deleting the cffi files until branch-0.6 is open. This is so that 0.5 includes the code in case that a binding bug would escape us, a fast hotfix to roll back to cffi can be done for 0.5 for fast bug resolution then. 0.6 then should be fully cffi free (queue party sounds).