[Python] to_pandas drops the time zone of a dictionary-encoded timestamp column #55490

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

to_pandas() drops the time zone of a dictionary-encoded timestamp column. The column becomes a pandas category whose categories are tz-naive timestamps. Comparing it with a tz-aware timestamp then returns False for every row, with no error:

import pandas as pd
import pyarrow as pa

ts = pa.array([0, 0, 10**9], type=pa.timestamp("ns", tz="UTC"))
plain = pa.table({"t": ts}).to_pandas()["t"]
encoded = pa.table({"t": ts.dictionary_encode()}).to_pandas()["t"]

print(plain.dtype)                       # datetime64[ns, UTC]
print(encoded.cat.categories.dtype)      # datetime64[ns], time zone dropped

t0 = pd.Timestamp(0, tz="UTC")
print((plain == t0).tolist())            # [True, True, False]
print((encoded == t0).tolist())          # [False, False, False]

The categories should keep the time zone, as datetime64[ns, UTC]. encoded > t0 and encoded.dt.tz_convert(...) also raise TypeError, while both work on plain.

Version: pyarrow 26.0.1, pandas 3.1.1, Python 3.15, macOS.

Component(s)

Python