Consider removing arguments that are irrelevant for certain models

[joshuacwnewton]

In much the same way we add arguments that are only relevant for specific models (e.g. https://github.com/spinalcordtoolbox/spinalcordtoolbox/pull/4860/commits/dbbbd87d8fe9a82946ff41c237e0098551e07906), we could consider removing arguments that have no meaningful effect on the output for certain models.

For example, for multiclass models (especially that of something like totalspineseg), options such as -thr and -keep-largest don't exactly make sense, do they?

That way, the interfaces for each model remain simple, and options are only provided if they are useful in some way or another.

[valosekj]

Good idea!

For example, for multiclass models (especially that of something like totalspineseg), options such as -thr and -keep-largest don't exactly make sense, do they?

Same for the rootlets model (rootlets_t2).

[SomeoneInParticular]

Agree; might it be worth making SCTArgumentParser have a subclass specifically for DeepSeg then?