Offer config blocks and deferred declarations to importers

[yenda]

Fixes #89.

Bug

The offered-names set that a qualified reference alias/Name is validated against (collect_referenced_trigger_names) held declared type names and referenced triggers only. Two reference forms drew false allium.reference.unknownName warnings (exit 1) although the aliased module visibly carries the construct:

1. alias/config.param — contradicting the language reference. The reference's "Config parameter references" section documents exactly this form:

A config parameter's default value can reference a parameter from an imported module's config block using a qualified name: required_copies: Integer = core/config.required_copies

and checker rule 46 requires it to resolve ("A qualified config reference in a default expression must resolve to a declared parameter in an imported module's config block"). On main, even that documented config-default position warns — and the message claims the module "does not define" a config block it plainly defines.

2. alias/DeferredName — a deferred declaration the module carries. Natural cross-module form: a module declaring deferred UnhandledException is exactly the module another spec wants to name in requires: mobile_auth/UnhandledException(request).

Repro (fails on main, passes here — transcribed into cross_module_lifecycle.rs): provider with a config block + deferred ExternalHelper; consumer referencing p/config.page_size (both in a config default and in a rule) and p/ExternalHelper → 3 false warnings on main, 0 here.

Fix

collect_referenced_trigger_names additionally offers:

• config, when the module declares a config block;
• each deferred declaration's root name (deferred Foo and deferred Foo.bar both offer Foo). A deferred declared against another module's alias (deferred other/Foo) names a construct that module owns and is deliberately not offered — extract_leading_ident returns None for qualified paths.

Not in scope (follow-up)

Field-level validation of config references (rules 46/47: does page_size exist in the imported block, do the types match). Today the QRef for alias/config.page_size carries only the name config, so a field typo produces the byte-identical diagnostic to a valid reference. This fix stops the false positive on the block reference; field resolution needs the qref collector to carry the member chain.

Tests

• 3 unit tests on the offering set (config + deferred + dotted-deferred root offered; neither offered by a module without them; a foreign deferred other/Foreign offers neither the name nor the alias)
• 3 integration tests (the ticket repro resolves and exits 0; p/NoSuchDeferred still warns; p/config against a module with no config block still warns)
• Full workspace suite green: 655 tests, 0 failures.

Real-world impact

On a ~180-spec production corpus, the v3.5.0 → v3.5.3 upgrade produced 8 false unknownName warnings of these two shapes (5 config, 3 deferred), each now carrying a suppression directive plus an explanatory comment. With this fix those suppressions become unnecessary.

---

Note: this PR and the config/deferred one both append tests to cross_module_lifecycle.rs and the analysis.rs test module — whichever merges second needs a trivial union rebase; happy to rebase promptly.