The dependency graph already encodes the information about base document field dependencies, but it is not directly queryable. 

We can likely re-implement Pipeline::getDependencies using the graph. We can probably use logic similar to getDeadFields to gather these, but will need to consider and track metadata too. This refactor can help uncover bugs and reduce tech debt.

If we can re-use the existing copy of the graph, this can also save us the cost of repeated pipeline deps analysis (performance).

We should only do this if it actually reduces code complexity and can be done in a self-contained way.