`hinfnorm` needs tweaking

[baggepinnen]

hinfnorm fails occasionally and it would be nice to come up with better default values for the tolerances etc.

success = map(1:10000) do i
    G = ssrand(1,1,4)
    try 
        hinfnorm(G)
        true
    catch
        @show poles(G), tzeros(G)
        false
    end
end
mean(success)

it fails in about 0.5% of these random cases, which is a bit much to rely on in CI etc.