Pandas 2.x support

[joe-clickhouse]

Fortunately, there's a relatively small surface area of pandas we need to be concerned with. In looking at the backwards incompatible API changes, most of the changes that'll concern us revolve around datetime64 & timedelta64 dtype resolutions.

Things we'll need to consider

1. Default use should be consistent with pandas 1.x behavior for backward compatibility and unsurprising differences between using 1.x vs 2.x:
   • Pandas 2.x supports resolutions “s”, “ms”, “us”, and “ns”. 1.x did not. In 1.x when constructing a Series or DataFrame and passing a datetime64 or timedelta64 dtype with unsupported resolution (i.e. anything other than “ns”), pandas would silently replace the given dtype with its nanosecond analogue. Captured in Using pandas 2.0 cause query_df to return columns with different/unexpected dtypes #165
2. Will need to incorporate running the relevant tests in CI against pandas 1.x and 2.x
   • Probably 2 additional tests: (Python 3.13, ClickHouse latest, pandas 1.x) and then (Python 3.13, ClickHouse latest, pandas 2.x)
   • See this comment for more context, but we'll need to add one new test: (Python 3.11, ClickHouse latest, pandas 1.x). Captured in Add pandas 1.x and 2.x testing to CI #528

Useful enhancements once supported

• Leverage pandas 2.x support of "pyarrow" dtype_backend: Support PyArrow backed Pandas DataFrames. #326
• We should expose a setting in common that'll allow users of 2.x to opt in to the more appropriate resolutions (e.g. [s] for Time, [us] for Time64(6)) or [D] for Date, etc.): Expose setting allowing pandas 2.x datetime64/timedelta64 variable resolutions #531