Thank you for the honest answer. One option I see for now is to interface with NetworkX. If you are familiar with it, do you have a sense of how terrible of an overhead I can expect by ferrying such deletions and Steiner tree queries across the language barrier?
Alternatively, I’m willing to look at how difficult it would be to adapt steiner_tree to return a mapping like you suggest.
Out of curiosity, what kind of funding supports development like that? Institutional? Funding agency?