feat(runtime): apply nested redefinition chains below composite features

[devin-ai-integration]

What and why

A chain redefinition written on a usage — part top : Top { attribute :>> mid.leaf.value = 99.0; } — is legal SysML v2 and resolves cleanly, but neither the pinned pilot nor OpenSysML gave it any effect: top.mid.leaf.value still read the definition default. Modelers were writing the shorthand and silently getting nothing. Only the nested-body form part :>> mid { part :>> leaf { attribute :>> value = 99.0; } } worked.

This makes the shorthand behave exactly like the nested-body form:

• semantics.NestedRedefinitionsOf(sym) (side table) lists the members of sym whose :>> target is a ≥2-segment feature chain that resolves.
• Runtime: when materialize(sym, id, owner, feature) creates a nested object, it collects the chains of owner's types (and their member sources) whose first segment is feature (or any alias sharing its *FeatureValue), plus tails carried on owner. A one-segment tail replaces that feature's EffectiveFeature on a cloned shape with effectiveFeature(name, redefiningMember, sym) — so value (=/default =), declared type, multiplicity and a redefining body all take effect and CompositeTypeOf picks them up unchanged; longer tails are stored on Instance.nested and applied when the next level materializes. materializeMembers goes through the same path, so a multi-valued intermediate (:>> wheels.radius) applies per element.
• Precedence is exactly the nested-body form's. Each chain is attributed to the type or usage whose body declares it (redefinitionContext: the enclosing definition, or the topmost usage). Between chains, a chain whose context specializes another's wins (so a classifier Sport :> Base added later beats Base's chain, for children materialized before or after the classification); carried tails and unrelated contexts keep first-wins. Against a standard redefinition already standing on the feature, the chain yields only when that redefinition's context conforms to the chain's — the same body, or a subtype's nested body (blocksChain); the child's type's own :>> and bodies in types the chain owner specializes lose, as they do to part :>> wheel { :>> radius = 3.0 }.
• Classification (classify) applies the new classifier's chains to already-materialized descendants transactionally (installFeatureValue) and leaves them pending for lazy ones; it only walks objects the parent owns, so a populated ref/port/subject never has its target mutated. Held images (Image/Restore) round-trip Instance.nested.
• Chains below a feature bound to an existing object follow the nested-body rules: a valued chain written in the same body as the binding (part :>> mid = existing; attribute :>> mid.leaf.value = 99.0;) is rejected with the existing ErrValuedFeatureRestated; a valued chain from a body that specializes the binding's (Top { part mid : Mid = existing; }, Sport :> Top { attribute :>> mid.leaf.value = 99.0; }) governs the inherited binding — EffectiveFeature.GovernedByChain makes valueBinds false, so a fresh object is materialized (also on classify) and the bound object is never touched; a chain declaring only a type or multiplicity conflicts with nothing.
• Expressions are evaluated in the declaring body's scope (DefaultDecl = the redefining member), so :>> mid.leaf.value = factor * 2.0 reads factor from Top, as the nested-body form does.
• New NestedRedefinitionPass reports an error, redefinition-through-reference, when a non-final segment is a ref/port/subject — there is no owned object to redefine below it, so the chain would silently not apply. A plain chain gets no diagnostic.
• The = vs default = fixed-binding rule is unchanged and still applies to the chain's last feature.

Specification basis

A feature chain in a redefinition target parses to a feature that hosts the chain; the chain determines the host's featuring type (first segment) and featured type (last segment), and the host is redefinable — KerML 1.0 §7.3.4 (feature chains) and §8.3.3.3 (redefinition), as confirmed with the spec lead. Applying the redefinition below the chain is therefore the specification's reading, recorded as ✅ Faithful in docs/project/spec-compliance.md (redefinition section) and described under "Chain redefinitions" in docs/reference/grammar/conformance-audit.md. The pinned pilot evaluator (0.62.0) accepts the notation but reads the original value; that is noted as a pilot-evaluator gap in docs/project/pilot-differential.md. No corpus ratchet or differential baseline moves.

How it was verified

• Conformance fixtures nested_redefinition_chain (3- and 2-level chains, chain with a body, outer-scope value, collection intermediate), nested_redefinition_chain_equiv (shorthand and nested-body forms read identically) and nested_redefinition_precedence (subtype over base, carried over intermediate).
• TestRuntimeRobustnessNestedRedefinition: chain through a ref evaluates to the default without panic; unresolvable chain returns ErrNoSuchFeature; chain below a value-bound feature returns ErrValuedFeatureRestated and leaves the bound object untouched.
• Unit tests for NestedRedefinitionsOf (1-level excluded, 2/3-level included, unresolved excluded), the pass (ref error; silent on plain chains and the standard forms, in default and strict mode), aliased names in both read orders, held-image round-trip of pending tails, and classify: chains reach materialized and lazy children, a later classifier's chain outranks the base type's (both read orders), the chain outranks the child's type's own redefinition while a subtype's nested body keeps its value, and a chain through a ref leaves the referenced object alone.
• By hand with bin/sysml REPL (recording in the session): shorthand and nested-body forms both → 99.0 / 4.0, outer-scope expression → 14.0, collection → [0.4, 0.4], 4-level chain, inherited chains, fixed-binding protection, ref/port/subject errors.
• gofmt -l ., go vet ./..., make lint, make docs-check, go test ./... all clean.

Checklist

• make test and make lint pass locally
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved (compliance rows need nothing: the census is counted at docs build)
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/14f60f720c2d4586b28ed46025fa9f83
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/14f60f720c2d4586b28ed46025fa9f83?variant=devin
Requested by: @HuiJun

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring