Instead of having the analyses working directly on the DAG, they should generate a set of patches for each pipeline variant
There are three high-level patch types: Data patches (data filter, data projection, data transformer), Model patches, and Pipeline patches (delete node, replace node, and add node for the extraction nodes)
These patches should have a boolean attribute to decide if they require later parts of the DAG to be recomputed. This way, we can handle the extraction nodes also as pipeline patches and do optimisations later
Description