[✨ Feature Request]: Preprocessing

[joaquincabezas]

Is your feature request related to a problem? Please describe.

As discussed in #47 I was thinking on a previous stage where we can be sure that some logic is applied just to the facts.

Describe the solution you'd like

As Gus posed it: "prune out the preprocessing subgraph (not just a chain as it is now), and make its end nodes the new fact neurons (hence effectively skipping it in training)."

Describe alternatives you've considered

Doing this outside the logic language, but was dissapointing

Additional context

No response