Attempts to fix #27, currently fails due to keyword arguments in interp_kind

[dncnwtts]

Including the new scipy API for cubic interpolation has brought to light certain subtleties in how interpolation should be done for albedos and phase functions at scattering wavelengths.

[MetinSa]

I was discussing the interpolation of just three points with @eirikgje and @unfunfunt and what was suggested was that we use linear interpolation with for instance np.interp for interpolation and for extrapolation we just pick the edge values. Thoughts @dncnwtts ?

[dncnwtts]

I would like something like this:

Linear interpretation:
Fit lines between all non-zero points, and set to constant outside

Cubic interpolation:
Fit cubic splines between all non-zero points, but allow for extrapolation beyond that. The problem right now is that there are only three nonzero points, so this would be used only when we have more bands.

What do you think of this?

[hermda02]

Seems reasonable and easy enough to implement. Over a certain number of bands, a cubic spine is used. Perhaps some light testing against simulated data to determine at what point we should switch from linear to cubic.

[MetinSa]

Seems reasonable and easy enough to implement. Over a certain number of bands, a cubic spine is used. Perhaps some light testing against simulated data to determine at what point we should switch from linear to cubic.

As in, if a model parameter is tabulated for more than N frequencies, we use cubic spline?

Btw @dncnwtts regarding the use of np.nan instead of 0 for albedoes and phase coefficients: Tried this in #29 and forgot that this then means that there is a non-zero scattering contribution at all wavelenghts, which does not reproduce the K98 model simulations. I think we just set it to 0 like in the K98 model.