Fix Nyquist indentation geometry and automatic direction selection - #1265

Follow-up to #1223 for #1194 and #1191.

The explicit-direction fix still reflects the right-hand offset when indenting left. For example:

ct.nyquist_response(
    ct.tf([1], [1, .05]), indent_radius=.1, indent_direction='left')

On current main this places the contour on the pole and raises RuntimeError: attempt to evaluate at a pole. The corrected contour starts at -0.15, exactly 0.1 from the pole.

This change:

• makes None the automatic mode, preserving the previous default side selection (right of stable/on-axis poles, left of unstable poles);
• applies explicit left/right choices consistently through keywords and configuration, including an explicit None override;
• centers both indentation directions on the pole and samples the full connecting circular arc when forcing a detour across an off-axis pole;
• uses the poles enclosed by the indented contour for the encirclement check and retains warnings when a detour can cross a closed-loop pole.

Regression tests cover continuous/discrete systems, poles at the origin and at nonzero imaginary frequencies, either sign of a tiny real part, distances inside/on/outside the radius, automatic/manual frequency selection, configuration precedence, and the original #1191 transfer function in TF and state-space form. I also reproduced the ignored-direction behavior using the 0.10.2 response code.

Validation (Python 3.13.1, NumPy 2.2.5, SciPy 1.16.1, Matplotlib 3.10.1, Slycot installed):

• MPLBACKEND=Agg python -m pytest control/tests/nyquist_test.py control/tests/freqplot_test.py control/tests/response_test.py control/tests/config_test.py control/tests/docstrings_test.py -q: 435 passed, 8 skipped, 1 xfailed.
• MPLBACKEND=Agg python -m pytest control/tests --ignore=control/tests/optimal_test.py -q --timeout=60: 4577 passed, 51 skipped, 6 xfailed.
• Ruff, compilation, and git diff --check passed.

The unrestricted suite hit the 60-second timeout in optimal_test.py::test_optimal_doc, in SciPy optimization/integration; the optimization module was excluded from the broader completed run.

AI disclosure: OpenAI Codex generated the implementation, tests, and documentation, investigated and reproduced the bugs, and ran the validation commands. This PR is submitted at the account owner's explicit request.