Code deduplication

[dxqb]

This is an attempt to de-duplicate identical code between models,

• but keep the OneTrainer object model unchanged
• still have the possibility to write model-specific classes if that is necessary for future models
• this makes changes easier for future PRs. as an example, one of the existing PRs requires passing quantization settings through the model loader

This PR is only model loaders, but the principle could be applied to other parts of the object model.
This PR uses class constructors, similar to C++ templates. An alternative to this way would be to use base classes and mixins, but I feel this way is cleaner and less likely to break things. Feedback welcome

[Nerogar]

I haven't tested the change, but it's much shorter and still readable. I usually prefer inheritance and mixins. But in this case I think it's ok to deviate from the usual style if it makes the code easier to maintain.

[dxqb]

Thanks! It has been merged into #1034 which people have tested, so it should already be tested reasonably well.