fix(como_cmaes): default reference point from the known nadir, and report kernels when none dominates it

[sean-from-japan]

Fixes #811

Changes

• When reference_point is None and the problem knows its Pareto front, the reference point is now nadir + 0.1 * (nadir - ideal). COMO-CMA-ES approximates the optimal p-distribution of the hypervolume, so its reference point should be fixed just beyond the nadir, as in the authors' bbob-biobj benchmark. If the front is unknown, or loading it raises OSError (for example a failed download), the current far fallback is kept.
• If no kernel dominates the reference point, _advance now reports the kernels' incumbents. Before, self.pop was never updated in that case and the result was the initial random sample. This showed up on ZDT4 once the reference point was near the nadir.
• Updated the reference_point docstring.
• Added four tests.

Results

Default arguments, 300 generations, seeds 1–3:

Problem	main: front size	main: IGD	this PR: front size	this PR: IGD
ZDT1	8–10	0.099–0.107	32	0.012
ZDT2	2–3	0.351	29–32	0.017
ZDT3	10–11	0.12–0.24	14–21	0.030–0.045
ZDT4	5–9	12–20	4–8	2.2–3.6
ZDT6	5–11	0.29–0.30	15–20	0.067–0.109

Problems without a known Pareto front, such as bbob-biobj through cocoex, take the unchanged fallback, so the bbob-biobj behaviour measured in 527ec67 does not change.

Testing

• pytest tests/algorithms/test_como_cmaes.py tests/algorithms/test_mocmaes.py: 11 passed.
• On main, test_como_cmaes_default_reference_point_uses_known_nadir and test_como_cmaes_default_does_not_stall_on_zdt1 fail. Without the second change, test_como_cmaes_reports_kernels_when_none_dominates_reference_point fails.
• Merges cleanly with fix(como_cmaes): reject constrained problems instead of ignoring constraints #810, and the combined test file passes.
• No new ruff check or ruff format --check findings.

Note

problem.pareto_front() downloads the front for some problems, and no other pymoo algorithm reads the known front during a run. If you prefer to avoid that, I can reduce this PR to the reporting fix and the documentation.