Add support for pandas-2.2 in cudf

[galipremsagar]

Description

This PR:

• Enables pandas-2.2 in cudf by upgrading the upper bound pinnings.
• Cleans up a lot of dead-code.

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]

Just in noting in the future we should pass pandas a numpy array instead of a pyarrow array and pandas.Series treats pyarrow arrays as objects

[mroeschke]

If/when we decide to test against the minimum supported pandas version (pandas 2.0) would we have to add any of these checks back in?

[galipremsagar]

If/when we decide to test against the minimum supported pandas version (pandas 2.0) would we have to add any of these checks back in?

The current PANDAS_GE_200 code deletions will also hold if we want to test with pandas-2.0 (the lowest supported version), but I think it's highly unlikely we can keep up and maintain such a test suite in CI. Most of the checks being removed in this PR were gating us to distinguish test code from pre-2.0 vs post 2.0 behavior.

[shwina]

Is this the right pinning, or should we go with <2.3.0dev0? Do we expect patch releases to break us?

[galipremsagar]

I feel various fixes are going into 2.2.1 that may have an impact on us, but would defer to @mroeschke on that.

[mroeschke]

Yeah I fixed various issues @galipremsagar found in pandas 2.2.1 (coming out next week) which there are xfails for so the update will break the test suite. We should probably relax the pin in a follow up PR when that's out

[shwina]

Approving with a question about pinnings that we should resolve before merging.

[galipremsagar]

/merge

[galipremsagar]

@shwina @raydouglass @mroeschke In the interest of time, I upgraded the pinnings to include the latest patch release of pandas-2.2.1 too. If you have any concerns on these changes, let me know I'll address it in a follow-up pr.