Stabilize LQE covariance symmetry checks - #1227

Summary

Fixes #1174.

This makes the Riccati symmetry checks more robust and avoids false QN must be a symmetric matrix errors when lqe() / dlqe() form the projected process covariance G @ QN @ G.T.

Details

• Replace the local floating-point symmetry check with SciPy's issymmetric / ishermitian using a small scale-aware tolerance.
• Validate user-supplied QN as a symmetric covariance matrix before projection.
• Symmetrize only the projected covariance G @ QN @ G.T, which is mathematically symmetric but can pick up roundoff-level asymmetry.
• Add deterministic regressions for both lqe(..., method="scipy") and dlqe(..., method="scipy").
• Add a guard test showing clearly asymmetric user QN is still rejected instead of silently repaired.

Local validation

• Red proof: using the regression matrices without the projected-covariance symmetrization makes care() and dare() reject G @ QN @ G.T with ControlArgument: QN must be a symmetric matrix; the measured projected asymmetry was 4.44e-16.
• python -m pytest -p no:cacheprovider control\tests\mateqn_test.py::TestMatrixEquations::test_symmetric_shape_check_tolerance -q
• python -m pytest -p no:cacheprovider control\tests\stochsys_test.py::test_lqe_symmetrizes_projected_process_covariance control\tests\stochsys_test.py::test_dlqe_symmetrizes_projected_process_covariance control\tests\stochsys_test.py::test_lqe_rejects_user_supplied_asymmetric_covariance -q
• python -m pytest -p no:cacheprovider control\tests\mateqn_test.py -q -> 11 passed, 13 skipped (slycot not installed)
• python -m pytest -p no:cacheprovider control\tests\stochsys_test.py -q -> 29 passed, 2 skipped (slycot not installed)
• python -m ruff check --no-cache control\mateqn.py control\stochsys.py control\tests\mateqn_test.py control\tests\stochsys_test.py
• PYTHONPYCACHEPREFIX=C:\tmp\pycache-python-control-1174 python -m compileall -q control\mateqn.py control\stochsys.py control\tests\mateqn_test.py control\tests\stochsys_test.py
• git diff --cached --check

AI-assisted contribution disclosure: implementation prepared with OpenAI Codex and reviewed/tested locally before submission.

AI Disclosure: Codex (ChatGPT 5.5) was used during code navigation, initial drafting, and PR text preparation. All logic, code changes, and test cases have been manually reviewed, verified, and tested locally by the author in accordance with the NumPy AI Policy.