Problems with baltrunc

[olof3]

As reported by @ma-laforge in #298, there are some problems with baltrunc

A = diagm([-1.0, -2, -3])
B = [1 2; 3 4; 5 6]
C = [1 2 3]
D = [1 -1]
sys = ss(A, B, C, D)

julia> baltrunc(sys)
ERROR: The number of columns of B (2) and D (1) are not equal

The problem is the following line

if unitgain
    D = D/(C*inv(-A)*B)
end

The implementation of baltrunc does not seems consistent with the section on model reduction in the given reference (Glad & Ljung), so I am not sure what the fix would be. I guess one solution would be just to update the whole function according to the reference. Do you remember anything about this @baggepinnen

[baggepinnen]

Hmm, can't remember, seems it was some 5 years ago :P I might have implemented balreal from the book but taken a shortcut on baltrunc. I have a slightly more general implementation here
https://github.com/JuliaControl/RobustAndOptimalControl.jl/blob/24f974452cf12eaffa7fe0e9d9376edfe5f17ab4/src/reduction.jl#L8
that also supports frequency weights, perhaps we could generalize baltrunc to allow those weights, and default to weights 1. This implementation does not consider the static gain either, so that has to be added no matter what implementation is used.

[baggepinnen]

Actually, handling of static gain is a special case of weighted reduction, so it could potentially be handled by having the weight 1/s. It's not exactly the same as unweighted reduction with static gain compensation though.