fix: Clean up temp CSV in AthenaQuery.as_dataframe (v2)

[lucasjia-aws]

Issue

#5100

Problem

sagemaker.feature_store.feature_group.AthenaQuery.as_dataframe() downloads the Athena query result to <tempdir>/<execution_id>.csv, loads it with pd.read_csv, and never deletes the file. Each query leaves a CSV behind, so running many queries (or queries with large results) can fill up local disk space.

Root cause

as_dataframe returns pd.read_csv(filepath_or_buffer=output_filename, ...) directly with no cleanup. DatasetBuilder.to_dataframe in the same package already removes its downloaded file after reading; as_dataframe did not.

Fix

• Wrap the download and read_csv in try/finally and remove the temporary file in finally, so it is deleted after a successful read and also when the download or parsing raises.
• Cleanup is best-effort via a small _remove_temp_file helper: it only removes the file if it exists, and an OSError (for example on Windows when a chunksize/iterator reader still holds the file open) is logged as a warning instead of failing the call.
• Docstring notes that the temporary file is removed.

Testing

• Added unit tests in tests/unit/sagemaker/feature_store/test_feature_group.py: the temp file is removed after a successful load (using a real temp directory), it is removed when read_csv raises, and a failing os.remove does not break the call.
• The two new cleanup tests fail without the fix and pass with it.
• pytest tests/unit/sagemaker/feature_store/test_feature_group.py: 56 passed.
• black --check and flake8 pass on the changed files.

V3 counterpart: #6344