test ag08bd agnostic of Af-lambda*Ef scaling

[bnavigator]

Fixes #157

From the AG08BD.f docstring:

C     PURPOSE
C
C     To extract from the system pencil
C
C                       ( A-lambda*E B )
C           S(lambda) = (              )
C                       (      C     D )
C
C     a regular pencil Af-lambda*Ef which has the finite Smith zeros of
C     S(lambda) as generalized eigenvalues. The routine also computes
C     the orders of the infinite Smith zeros and determines the singular
C     and infinite Kronecker structure of system pencil, i.e., the right
C     and left Kronecker indices, and the multiplicities of infinite
C     eigenvalues.

As reported in #157, some math libraries out there compute Af, Ef with a different scaling as the SLICOT example. The result should still be valid.

[bnavigator]

I verified this on Arch Linux, which failed before: https://build.opensuse.org/package/show/home:bnavigator:archlinux/python-slycot

[roryyorke]

Looks good to me; from Wikipedia the generalized eigenvalue problem is Af @ x == lambda * Ef @ x, so Af and Ef are not unique.

This change can't be causing the test failures. I see control/tests/flatsys_test.py::TestFlatSys::test_point_to_point_errors is common to all the failing runs, and that two cases fail control/tests/timeresp_test.py::TestTimeresp::test_forced_response_T_U[dt=True, no T] and control/tests/timeresp_test.py::TestTimeresp::test_forced_response_T_U[dt=0.2, no T]. I don't have any suggestions for this, except the not-so-helpful one of ignoring the failures and merging anyway.

[bnavigator]

Thanks.

The mentioned failures should be fixed by python-control/python-control#659. I just retriggered the CI jobs now. If it succeeds, all is well. If not, we can still ignore.