PandasCursor and ArrowCursor fail on columns with the same name and different types in the S3 result file

[laughingman7743]

Problem

When columns with the same name have different Athena types, PandasCursor and ArrowCursor fail to read the S3 CSV result file, because the type maps they pass to the CSV reader are keyed by column name, and the reader applies an entry to every column with that name.

SELECT 1 AS x, 'a1' AS x, CAST('12:34:56' AS TIME) AS x, measured on Athena on 2026-10-04 with #1050 applied (8e5dda6):

Cursor	S3 result file	managed (GetQueryResults)
PandasCursor	OperationalError: Unable to parse string "12:34:56.000" at position 0	[(1, 'a1', time(12, 34, 56))]
ArrowCursor	ValueError: time data '1' does not match format '%H:%M:%S.%f'	[(1, 'a1', time(12, 34, 56))]
PolarsCursor	[(1, 'a1', time(12, 34, 56))]	[(1, 'a1', time(12, 34, 56))]

• pandas. pandas.read_csv(dtype={"x": ...}) applies the dtype to x and to the renamed x.1, x.2 (checked with pandas 3.0.6). The integer dtype of the first x is applied to the varchar and time columns.
• Arrow. pyarrow.csv.ConvertOptions(column_types={"x": ...}) types every column named x, and AthenaArrowResultSet.converters keeps one converter per name, so the converter of the last x is applied to all of them.

#1032 (PR #1050) fixed columns with the same name and the same type. It keeps converters and parse_dates per column under pandas' and Polars' renamed names, but it does not change how the pandas dtype and pyarrow column_types maps match these columns.

Before #1050, the same query also failed for PolarsCursor on the S3 path and for all three cursors on the managed path.

Expected

Columns with the same name and different types convert by their own types when read from the S3 result file, as they do through GetQueryResults and in Cursor.

Environment

• PyAthena master 28ec68d + Keep columns with the same name, and key CSV column types by the reader's column labels #1050 (8e5dda6), Python 3.13.1, pandas 3.0.6, pyarrow and polars from uv.lock.
• UNLOAD is not affected: Athena rejects duplicate column names there (DUPLICATE_COLUMN_NAME).

[laughingman7743]

Correction after PR #1050 4585ef7: the Arrow fetch converters are now chosen by column position, so ArrowCursor converts same-named columns with different types correctly on the managed path, and on the S3 path when they share an Arrow type (e.g. json and varchar). What remains for Arrow is the S3 read itself: pyarrow.csv column_types is keyed by name, so different Arrow types (e.g. integer and time) still fail.