Fix overflow case and clean up some logic

[vyasr]

Description

The calculation of the size in bytes for the CUDA Array Interface for pylibcudf Column objects produced from a column_view and an arbitrary owner previously ran the risk of overflow because the arithmetic was performed on int32 types but that is actually the maximum size in number of elements, not bytes. Since the CAI is a Python object, we can do the arithmetic with pure Python (infinite precision) integers to avoid this problem. In the process of fixing this bug, this PR also does some minor cleanup of the various cases handled in the size calculation.

Resolves #18598

Checklist

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

[mroeschke]

I suppose it's onerous to create a large enough column to test that we don't overflow?

[Matt711]

Should we add the reproducer (or similar) in the issue to our test suite?

[vyasr]

Yeah I didn't include a repro because the test in #18598 takes 1000 s to run. The big issue is that the assertion takes forever since it requires converting to a pandas object and checking on the CPU.

[vyasr]

/merge