Benefits Denial's logistic regression also stops at max_iter #1237

Where: faircode/models.py, results/.

Current gap: #1152 names German Credit Lending and Healthcare Readmission, but a fit of the baseline and unawareness feature sets of Benefits Denial also raises ConvergenceWarning (the other three strategies do not), so the baseline rows of that audit depend on where the solver stopped.

Repro:

python3 -m pytest tests/test_benchmark_convergence.py -q -k 'Benefits'

Suggested approach: Decide with the two audits in #1152 whether to raise max_iter or scale inside the estimator, then rerun and record the movement with diff_results.py.

Done when: no benchmark fit of Benefits Denial warns and the entry in tests/test_benchmark_convergence.py is removed.