Compute the whole closure of to-be-executed generators upfront before executing anything
Properly compute closures for the 4 different scenarios:
1. full_closure: run all generators for a selected machine in topological order
2. all_missing_closure: run just the missing generators including their dependents
3. requested_closure: run only a selected list of generators including their missing dependencies and their dependents
4. minimal_closure: Run just enough to ensure that the list of selected generators are in a consistent state. Don't execute anything if nothing is missing.
