Stop a closed PolarsDataFrameIterator for every reader kind

[laughingman7743]

WHAT

PolarsDataFrameIterator.close() replaces its reader with an empty iterator before closing a generator reader, as PandasDataFrameIterator.close() does.
A closed iterator over a whole DataFrame (without chunksize, with managed query result storage, or for failed or empty results) now yields nothing and no longer holds the DataFrame; closing a chunk generator behaves as before.

Release note (4.0.0, fix): a closed PolarsDataFrameIterator, such as one returned by iter_chunks(), yields nothing for every result kind.

WHY

Closes #1019.

close() only handled GeneratorType readers, so iter([df]) kept yielding after close().

TEST

Tested commit: c7846e6

• just lint: passed.
• New offline test TestPolarsDataFrameIterator::test_close_stops_iteration (whole DataFrame and generator readers): uv run --env-file <dummy env> pytest --noconftest tests/pyathena/polars/test_result_set.py: 4 passed. With the source change reverted, the dataframe case fails and the generator case passes.
• uv run --env-file .env pytest -n 1 tests/pyathena/polars/test_cursor.py -k "close or iter_chunks or whole_result": 9 passed.
• Not run locally: the full just test pyathena suite; it runs in CI when the PR is marked Ready.

🤖 Generated with Claude Code

[laughingman7743]

Self-review round one (implementation behavior): CLEAN

Base de8cc52ac2a43cdba72bb4571c185883287741fc, head c7846e619459926f07cb5db6c1a342658bea2f54.

Covered: PolarsDataFrameIterator.close(), __next__() (which calls close() on StopIteration), __exit__, iterrows(), and the callers AthenaPolarsResultSet.close() and iter_chunks() (whole-DataFrame and chunk-generator readers).

• Whole DataFrame: after close(), the reader is iter(()), so iteration ends and the iter([df]) list iterator (with the DataFrame) is released.
• Generator: the generator is still closed (GeneratorExit at its yield; _iter_*_chunks catches only Exception), and the reader is iter(()) instead of the closed generator. Both end iteration the same way.
• A repeated close() and close() from __next__ after exhaustion are no-ops.
• The new offline test fails on the base for the dataframe reader. The generator case guards the existing behavior.

[laughingman7743]

Self-review round two (claims, callers, operations): CLEAN

Base de8cc52ac2a43cdba72bb4571c185883287741fc, head c7846e619459926f07cb5db6c1a342658bea2f54. Full pass over the PR body, the commit message, and #1019.

• "as PandasDataFrameIterator.close() does": it assigns self._reader = iter(()) before closing a TextFileReader.
• "whole DataFrame without chunksize, with managed query result storage, or for failed or empty results": AthenaPolarsResultSet.__init__ wraps self._df in those cases (since Keep the whole pandas and Polars results reusable #1005).
• Callers: no signature change. AthenaPolarsResultSet.close() (Keep the whole pandas and Polars results reusable #1005) calls self._df_iter.close() and then replaces the iterator, so its behavior does not change. No AWS requests are involved.

[laughingman7743]

Independent review (relayed): CLEAN

Reviewer: Codex CLI 0.160.0, model gpt-6-astra, reasoning effort high, sandbox read-only, session 01a1020d-2122-7b00-887a-6cae10cf0584. Static review only. Snapshot: detached worktree at head c7846e619459926f07cb5db6c1a342658bea2f54, base de8cc52ac2a43cdba72bb4571c185883287741fc. The prompt contained the literal diff and the intended behavior only. Afterwards, the snapshot and the PR worktree were clean.

Covered (reviewer): both reader kinds; __next__, repeated close(), __exit__, and iterrows(); result-set close() and iter_chunks(); the sync, async, and aio Polars cursors; CSV and Parquet generator cleanup (GeneratorExit bypasses except Exception); pandas parity; and the new test, whose dataframe case fails on the base. No regressions.

Pre-existing boundaries noted by the reviewer, not changed:

• A partly consumed iterrows() generator can still yield the rest of its current frame after close(); a new iterrows() yields nothing.
• A whole-DataFrame iterator returned by iter_chunks() is independent, so closing the result set does not close it (the Keep the whole pandas and Polars results reusable #1005 design). Closing it explicitly now works.