[_cffi_backend] Improve CField, CType, _CDataBase, buffer

[mildsunrise]

these were verified against implementation code. the docs are not very precise on some of these things, but I think it's clear all this is intended to be public facing API. CType in particular will be documented better as part of python-cffi/cffi#231.

draft because I still need to resolve the style issues (so far I can only see (1) replacing Optional with | None, and (2) importing Self from _typing_extensions).

I also need to see if Python/typeshed allows structural typing (because available CType fields vary depending on the value of kind)

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[srittau]

LGTM, I'll merge this when you undraft.

[mildsunrise]

@srittau the CType class exposes different sets of attributes depending on the kind. would it be acceptable to make it a union of protocols instead of a (nominal) class, like i'm doing here? the section on subtyping seems to say so, but this would break uses of isinstance(x, CType). however i feel like it would be more accurate / useful overall

[srittau]

I suggest we merge this PR as is, as it's a clear improvement on the status quo. Then we can experiment with the protocol-based approach in a targeted PR. I'm especially interested in the primer output.

[brianschubert]

This has conflicts now, @mildsunrise can you take a look?