Checkpoint tests for derived architectures

[frostedoyster]

Description of the issue

Our derived architectures such as the LLPR or the new Classifier architecture in #917 use a checkpoint testing mechanism that forces us to replace the checkpoint every time the architecture that is used as their base for the tests (currently PET for both) is changed

[pfebrer]

Since this is not the only difference of these architectures, I think we should at some point somehow differentiate between "primary" and "secondary" architectures in the way we store them, test them and document them.