Type instability _bounds_and_features

[mfalt]

We have some horribly type unstable code in _bounds_and_features

[ronisbr]

I think the biggest problem is that zpkdata is also type unstable. I understand that it is because it can be Real or Complex. However, wouldn't it be better if everything is always treated as Complex but print to the user as Real when the poles and zeros are real? I think this can help to improve the performance of some computations.