Addresses #1173.

This keeps the existing dlqe default behavior unchanged: by default it still returns the one-step predictor gain A P C^T (C P C^T + R_N)^-1.

The change adds return_filter_form=True for callers that need the filter-form correction gain P C^T (C P C^T + R_N)^-1, which cannot be recovered from the default predictor gain when A is singular. The returned covariance is documented as the steady-state prediction/prior covariance used by both gain formulas.

I also corrected the dlqe Riccati equation documentation and the stochastic docs wording that referred to dlqr/lqr in the estimator section.

Tests run locally:

• python -m pytest control/tests/stochsys_test.py::test_DLQE_return_filter_form control/tests/stochsys_test.py::test_lqe_discrete -q
• python -m pytest control/tests/stochsys_test.py -q
• python -m pytest control/tests/docstrings_test.py -q
• python -m pytest control/tests/mateqn_test.py::TestMatrixEquations::test_dare -q
• python -m pytest control/tests/kwargs_test.py::test_unrecognized_kwargs -q -k "dlqe or lqe"
• python -m ruff check control/stochsys.py control/tests/stochsys_test.py
• python -m compileall -q control\stochsys.py control\tests\stochsys_test.py
• python -m numpydoc render control.stochsys.dlqe