[CT-2744] Refactor: entry point for cross-project ref

[jtcohen6]

After much spiking and iterating on how to handle "public models from an upstream project," we've landed on a really satisfying implementation (#7890): These are just models. We needn't add special-case logic (or duplicate boilerplate code) within dbt-core for handling these nodes: listing them, resolving references to them, (not) running them, etc.

This implementation gives us a much narrower entry point within dbt-core for the resolution of cross-project ref. In #6725 (reply in thread), I shared more context around our intent for this feature. It should be uniquely available for dbt Cloud (Enterprise), given the problem it's solving—collaboration across multiple teams, with dozens or hundreds of contributors. It is not a publicly committed interface within the dbt-core framework. And it will yield a functional outcome comparable to that of using packages, via a mechanism that's significantly better at scale.

In a separate discussion earlier this year (#7173), we shared our aspiration for a future in which dbt-core is a pluggable library: narrowly scoped to its core workflow, while also providing more flexible points for extensions. Now that the refactor above is complete—cleaner interfaces, more-precise entry point—we can experiment with putting all logic specific to cross-project ref within a separate plugin.

So, the scope of this issue is to finish consolidating the remnants of this logic within dbt-core; to provide ourselves with a single clean entry point; and to make a first attempt at implementing a plugin framework, on which we can keep iterating.