Build the pandas result-set tests through the constructor #1257

Summary

Part of #1239. After #1256, tests/pyathena/pandas/test_result_set.py builds AthenaPandasResultSet in two ways:

• _local_result_set() (added by Remove and rewrite unnecessary pandas cursor tests #1256) calls the constructor over local result files. The Athena and S3 clients are mocked, and the module's S3FileSystem is patched.
• _read_csv_result(), _unload_result_set(), and _read_unload_result() predate Remove and rewrite unnecessary pandas cursor tests #1256. They create the instance with __new__, set about ten private attributes, and patch private methods and properties (description, output_location, _get_content_length, _read_data_manifest, is_unload).

The second kind tests a result set that __init__ never builds. A change to the constructor's state, such as a new attribute that the reads use, can pass these tests and fail for callers. Two helpers also give future tests a choice of fakes; #1256's reviews asked to keep a single kind.

Proposal

Move the tests that use the bypass helpers onto _local_result_set(), mainly the json-column tests and test_iter_parquet_chunks_null_in_later_file, then delete the bypass helpers. Each moved test must still fail under its mutation. Some json tests read CSV text that is not quoted as Athena writes it; check that they read the same data through the constructor. _fetching_result_set() patches _create_dataframe_iterator to feed prepared chunks and can stay, because the fetch tests need chunk boundaries that a local file does not control.