Finding a path through the mapping network should be sensitive to neural and non-neural mappings

[roedoejet]

See @joanise's description here: #446 (review)

[joanise]

Currently, our shortest_path implementation does not support any subgraph restrictions or options.

When the neural option is not enabled, we want to be able to ignore edges that are neural when doing the shortest path calculation, and maybe also return a targetted error message when the only path that exists for the requested conversion is neural.

Options to consider:

• create two graphs, with and without the neural edges, and leave the shortest_path algorithm alone
• implement an edge filtering option to shortest_path
• something else altogether?