Vector Fitting: Bug fix for #677 and #702

[Vinc0110]

Fixes issue #677 (and #702).
Also a warning is printed in case the passivity enforcement did not succeed (when n_samples was too small).

The problem was an incorrect choice of the evaluation frequencies for the passivity evaluation.

Quote from the paper [1]:
First, a dense set of frequencies is determined from dc up to about 20% above the highest relevant frequency. This highest relevant frequency is the maximum of the highest crossing from a nonpassive to a passive region on one hand and the maximum frequency of interest on the other hand.

The second condition was missing in the code.

[1] T. Dhaene, D. Deschrijver and N. Stevens, "Efficient Algorithm for Passivity Enforcement of S-Parameter-
Based Macromodels," in IEEE Transactions on Microwave Theory and Techniques, vol. 57, no. 2, pp. 415-420,
Feb. 2009, DOI: 10.1109/TMTT.2008.2011201.

[Vinc0110]

So I fixed one part, but I'm not sure if this really fixes all the problems. It still seems a bit strange for certain setups. I need to check it some more.

[Vinc0110]

This PR ended up as a major revision of the passivity enforcement routine. It now works much better and the routine offers a bit more control with the new parameters n_samples and f_max. Issue #677 is solved for n_samples=500.

I'll leave it for now and will merge soon, if no one complains.

[Vinc0110]

Passivity assessment and enforcement now works for most of my test cases. Enforcement still sometimes fails if the fit is terribly non-passive, which is probably a natural limitation. A massive number of samples would be required, which quickly becomes impractical. A UserWarning is issued in those cases, suggesting to re-fit the model with fewer poles and/or without the constants.