Include neutral model details in DXSM metadata

[willdumm]

Currently, a default neutral model is used as the basis for DXSM model training, but the details of this model are not included in model metadata. This should be fixed, maybe adding the name of the pretrained model from netam.pretrained which was used.

Also, we should include any multihit correction, and eventually add those corrections to the pretrained models. Since a multihit model has only three free parameters, these could be included directly in DXSM model metadata as well.