terraform: Further generalization of the EvalContext scope handling

[apparentlymart]

This is a continuation of some restructuring started in #34571.

In those earlier commits my goal was to minimize breaking changes to the EvalContext API to churn existing code as little as possible.

However, the way I achieved that was to handle fully-expanded modules and partial-expanded modules as two completely separate codepaths, which means that all other code must statically decide which of the two cases it's dealing with even though there's not really any reason the EvalContext implementation couldn't support that being decided dynamically.

For the "deferred actions" work, in a future commit we'll need to introduce a new graph node type representing partial-expanded resources which must decide dynamically whether it's in a fully-expanded module scope or a partial-expanded module scope, because the treatment is quite different for a fully-expanded module containing a resource with unknown expansion vs. a resource that's inside a module that hasn't been expanded itself.

This is therefore the logical conclusion of the previous work, making "eval context scope" the primary way that we talk about where an eval context is doing its work, with a helper function for the common case where that's a fully-expanded ModuleInstance.

In particular, this includes a third variation of the family of interfaces that have method Path indicating the scope that the node should be evaluated in. graphNodeEvalContextScope.Path may return any non-nil evalContextScope value chosen dynamically at runtime. The existing interfaces that statically return addrs.ModuleInstance and addrs.PartialExpandedModule respectively remain as statically-typed alternatives for the common case, since the need to decide dynamically is (for now) limited only to nodes representing addrs.PartialExpandedResource addresses.

This has unfortunately now churned a little more code than I originally wanted to, but the helper functions and the retaining of the statically-typed node-scope-discovery interfaces has kept it as small as possible.

(This also continues our effort to gradually make the implementation details of package terraform be unexported, at the expense of a little inconsistency in the short term where these new fields/types are mixed in with exported fields/types. Hopefully we'll continue on this quest over time and eventually reach things being consistently unexported.)

---

The first (and, for now, only) implementation of graphNodeEvalContextScope will follow in a future PR, adding a new graph node representing what the addrs.PartialExpandedResource address type represents: either an unknown resource instance address inside a known module instance address, or an unknown resource instance address inside an unknown module instance address.

[alisdair]

This is hard for me to fully comprehend without a concrete implementation of graphNodeEvalContextScope, but it looks like a good direction as far as I can understand it!

[github-actions]

Reminder for the merging maintainer: if this is a user-visible change, please update the changelog on the appropriate release branch.

[github-actions]

I'm going to lock this pull request because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active contributions.
If you have found a problem that seems related to this change, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.