Unified layer for abstracting relationship information

[bpbond]

Right now the data system contains all the information necessary to figure out dependencies and relationships:

• For chunks, they report what data objects they need and what they produce. This information is available via the chunk_inputs() and chunk_outputs() functions.
• For data objects, they contain precursor information. This information is available after the driver runs, and it's also statically preserved in the new GCAM_DATA_MAP object (see dstrace fix #751 ).

Right now, the various parts of the system that use this information (dstrace; graph_chunks; the driver itself; @pralitp 's PR #660 for figuring out minimum call sequence) get this information in tabular form and figure it out for themselves. It would be a lot better to abstract away these data structures, and have the user-facing routines query functions for the relationship information, e.g. by a call to some precursors(x, all = c(TRUE, FALSE)) utility.

In other words, we'd like, following good OOP practice, to expose relationship information without exposing how it's encoded.

[bpbond]

See also the now-closed #830