BUG: groupby.agg with udf incorrectly inferring pyarrow timestamp dtype

[jbrockmendel]

# based on test_agg_tzaware_non_datetime_result
import pyarrow as pa
import pandas as pd

dtype = pd.ArrowDtype(pa.timestamp(unit="ns", tz="UTC"))
dti = pd.date_range("2012-01-01", periods=4, tz="UTC").astype(dtype)
df = pd.DataFrame({"a": [0, 0, 1, 1], "b": dti})

gb = df.groupby("a")

result = gb["b"].agg(lambda x: x.iloc[0].year)
expected = pd.Series([2012, 2012], name="b")
expected.index.name = "a"
tm.assert_series_equal(result, expected)  # <- raises

The agg call does dtype inference on the result by calling maybe_cast_pointwise_result, which has special-casing (which we've wanted to get rid of for a while) for pd.DatetimeTZDtype.

AFAICT the way to fix this is to make _from_sequence stricter xref #33254.