Keep Graph from becoming a bloated junk drawer

[bobmcwhirter]

Per discussion on one of @ctron 's PRs, the graph crate is growing large, murky, and ill-defined.

I think possibly a structure along the lines of the following might be useful:

Merge the write aspects of graph with ingestor and make it the ingest module.

This would help solve the problem of how we're testing graph currently, using ingestor-y type of things.
This module would be hopefully the primary write operational module when it comes to touching the database.

Split the read aspects of graph out, to another module, fetch or similar.

But first, determine what form that should really take. We can certainly all share the entity crate and its models. We should probably not replicate all of the FooContext objects that are useful for ensuring a valid walk while writing the graph.

Reading will probably be a significantly more-used aspect than writing, at runtime, and so optimizing the fetches makes sense, instead of walking from a found root and aggregating things.

For instance, the /api/v1/advisory/{sha256} endpoint currently fetches the advisory, and then walks further/deeper to the joined advisory-vulnerability and vulnerability. This ends up being a 1+N operation, which it does not have to be (https://www.sea-ql.org/SeaORM/docs/relation/data-loader/). Here, the hierarchic FooContext of graph/ingest crate is actually limiting towards optimizations.

---

Ultimately, all of this will result in the graph crate no longer existing as it does today, and no singular crate named graph at all.