Ensure comparisons with pyints and integer series always succeed

[seberg]

When Python integers are compared to a series of integers, the result can always be correctly defined no matter the values of the Python integer.

This was always a very mild issue. But with NumPy 2 behavior not upcasting the computation result type based on the value anymore, even things like:

cudf.Series([1, 2, 3], dtype="int8") < 1000

would fail.
(Similar paths could be taken for other integer scalars, but there would be mostly nice for performance.)

N.B. NumPy/pandas also support exact comparisons when mixing e.g. uint64 and int64. This is another rare exception that cudf currently does not support.

Closes gh-16282

Checklist

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

---

Just as a note for reviewers. The later operator checks, need other to be fully coerced, so unfortunately can't deal with all operator special paths at the start.

[mroeschke]

/ok to test

[mroeschke]

Would there be a similar issue with comparing pyfloats and integer or float series?

[seberg]

Would there be a similar issue with comparing pyfloats and integer or float series?

Yes and no: we (as in numpy/pandas) are usually content with normal promotion logic there (unlike Python, which corrects for overflows and rounding when comparing int and float scalars).

[mroeschke]

/merge

[jakirkham]

Thanks Sebastian and Matt! 🙏