Fixes #1058.

StateSpace(A, B, C, D) currently coerces every state-space matrix through np.array(..., dtype=float), which silently discards imaginary parts from complex-valued matrices and emits ComplexWarning. State-space operations already evaluate complex frequencies and NumPy can safely keep real matrices as floats, so this changes _ssmatrix() to preserve complex dtype only when the input is complex-valued.

The regression test covers complex A/B/C/D matrices and treats constructor warnings as errors, so the old ComplexWarning path fails directly.

Validation:

• Reproduced the issue on current main before the fix: complex matrices were stored as real matrices and emitted four ComplexWarnings.
• python -m pytest control/tests/statesp_test.py::TestStateSpace::test_constructor_complex_matrices control/tests/statesp_test.py::TestStateSpace::test_constructor_invalid -q
• python -m pytest control/tests/statesp_test.py -q
• python -m pytest control/tests/convert_test.py control/tests/type_conversion_test.py control/tests/xferfcn_test.py control/tests/bdalg_test.py -q
• python -m ruff check control/statesp.py control/tests/statesp_test.py
• python -m compileall -q control
• git diff --check

Note: I also attempted python -m pytest control/tests -q, but the local Windows run exceeded a 5-minute timeout before producing useful output.

Assisted-by: OpenAI Codex