pandas 3 str dtype: performance depends on whether pyarrow is installed

In pandas 3.1.0, the default string dtype is 'str' in both cases, but its storage is 'pyarrow' when pyarrow is importable and 'python' otherwise; check Series.dtype.storage. The two behave differently in benchmarks. A 100k-row itertuples loop over float columns took 21-23 ms. With an unused str column added, it took 67-68 ms with pyarrow storage and 38 ms with python storage (object dtype: 24-25 ms). So a 'pandas 3 strings are slower/faster' result from one environment may not transfer to an environment without pyarrow, or the other way round. Report dtype.storage alongside the timings, and pin pyarrow in CI if performance matters.