Core: generating shapes from *predicted* peaks
In order to generate a better shape model, we might try to "bootstrap" the shape model by:
- constructing a shape model from found peaks
- applying that model to predicted peaks
- constructing a shape model from the resulting predicted peaks
- applying that model to the same predicted peaks
If we do not integrate the predicted peaks after step 2, this will fail, and we end up with a shape model with 0 peaks. This is not the intended behaviour.