Clarify and standardize the model type and interaction scheme taxonomy

[rfloca]

Currently the app interface specification has mixed semantics.
The field “type”: is e.g. used to encode types of algorithms but also interaction schemes, or even concrete implementations. (see e.g. Radiology App models “deepgrow” and “segmentation_spleen”
That is problematic for a robust implementation in host applications.

First ideas:

1. Interaction types should be separated from “type”
2. The interactions field should be able to take multiple values, as models might support several interaction schemes.
3. The values of the interaction field should be an open set, as we don’t know which kind of new interaction paradigms may come. We should only define the encodings for interaction schemes we already know now and have well defined.
4. It should be better clarified what “type” should encode and if there are a kind of taxonomy that can be used.