Fix port recommendations

[liamhuber]

Node recommendations were easy because we just needed to check if a node could support the demands. Port recommendations are harder because we need to check if a node instance does support the demands, i.e. we need to check if the current workflow graph is a subset of all possible ontological workflow trees.

With the more complex surface energy example, I discovered that some of the transitive demands were not being passed in the old, simpler port recommendation. So here we really do the graph-to-graph comparison described above. It's working great, and you can even invalidate a bunch of upstream stuff on-the-fly by re-wiring it to the wrong sort of downstream input. However the already un-performant task of selecting a new node is even worse; from something like an 0.7s delay to something like a 1.5s delay. I want feature and functionality before efficiency, so let's keep this and make it more performant in a follow-up.

[github-actions]

👈 Launch a binder notebook on branch pyiron/ironflow/fix_port_recommendation