[Python] Roundtripping with from_pandas().to_pandas() changes timestamp/timedelta resolution

[phofl]

Describe the bug, including details regarding any error messages, version, and platform.

With pandas 2.0 we support more resolutions than nanosecond resolution. E.g. the following should preserve the reso:

import pandas as pd
import pyarrow as pa
df = DataFrame({"a": [np.datetime64("2022-01-01")]}, dtype="datetime64[s]")
result = pa.Table.from_pandas(df).to_pandas()
result.dtypes

a    datetime64[ns]
dtype: object

Expected:

a    datetime64[s]
dtype: object

Tried on the latest nightly build

Component(s)

Python

[jorisvandenbossche]

Yes, this is known, historically we have hardcoded nanoseconds, which we can relax if pandas>=2 is installed.

See #33321, going to close this as a duplicate

[jorisvandenbossche]

Duplicate of #33321