Clean up misc, unneeded pylibcudf.libcudf in cudf._lib

[mroeschke]

Description

• Removed ctypedef const scalar constscalar usage
• Use dtype_to_pylibcudf_type where appropriate
• Use pylibcudf enums instead of pylibcudf.libcudf types

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[Matt711]

Non blocking question

[Matt711]

I was curious why we need to do this? Should we doing it for other types of sources?

[mroeschke]

I'm guessing that for raw string like data we're passing in views of the data into these readers, and Python will tend to garbage collect the sources sooner than the views that are passed into the readers.

Should we doing it for other types of sources?

Ideally no, I think generally it's good to not unnecessarily hold a reference to source if not needed so that memory (albeit probably small) is cleaned up

[mroeschke]

/merge