TST: fix parquet test_pandas_timestamp_overflow_pyarrow test (#8733)

This test started failing on arrow master (noticed thanks to our nightly builds testing dask). The reason for the failure is that we started using the `types_mapper` keyword in ``pa.Array.to_pandas`` (previously this was ignored for arrays), but in this test it is passed a wrong values (it should be a callable, not a dict).

dask/dataframe/io/tests/test_parquet.py

@@ -3072,9 +3072,7 @@ def clamp_arrow_datetimes(cls, arrow_table: pa.Table) -> pa.Table:
 
                     original_type = col.type
 
-                    series: pd.Series = col.cast(pa.int64()).to_pandas(
-                        types_mapper={pa.int64(): pd.Int64Dtype}
-                    )
+                    series: pd.Series = col.cast(pa.int64()).to_pandas()
                     info = np.iinfo(np.dtype("int64"))
                     # constrain data to be within valid ranges
                     series.clip(