Fix issue #156

[fweber144]

This fixes issue #156.

[fweber144]

Thanks for merging this. Just for the sake of completeness: I forgot to mention the following remarks which might become relevant when support for additive models is extended:

• I merged the "population-level interaction smoothers" into the "interaction smoothers" because to me, they appeared to be the same.
• In the t2() (formerly te()) term, I omitted argument bs because of what is described in issue Documentation of formula for additive models #156. Btw, I think this argument bs can be omitted in the demonstration anyhow because its default is "cr" and that should be the typical use case. (Compared to cubic regression splines, thin plate regression splines don't have an advantage as marginal bases for tensor product smooths (see ?mgcv::smooth.terms) and they are computationally more expensive.)