Cleaner solution for additional fit arguments

[chillerb]

Feature

In situations when you want to register a custom model, which has additional parameters, but which are not supposed to be tuned, it would be nice to have either a way to:

• Register model arguments, but not expose them for tuning. As far as I know, registering model arguments currently always require a parameter generator like a dials function to be passed via func
• Explicitly register specific engine arguments, that show up in show_model_info.

Just to make it more explicit: Right now, I'm working with a bunch of models that require additional prior knowledge Z.
Personally, it would probably be most intuitive to pass the priors to the fit function, but that isn't compatible with fit.model_spec's interface.
Therefore, I am passing the priors to set_engine, but that feels a little bit suboptimal.
First of all, there does not seem to be a way to document those engine arguments, so the user has no way of knowing what to pass there.
Also, the priors are more or less independent of the engine's implementation, which would point more in the direction of a model argument.

Not really sure, what the best solution would be, right now, I'm still sticking with set_engine.

Reproducible Example

Sorry, I don't have the time for adding a reproducible example right now, maybe I will add one later.