RUE-1209: delete the vestigial named-method and destructor projections

[steveklabnik]

Fixes RUE-1209. Refs RUE-1204, RUE-1027.

The semantic dependency manifest projected two body-level dependency channels that production never populates. named_method_dependencies and named_destructor_dependencies are extended only inside analyze_function_bodies_lazy, the whole-program body path; compose_queried_bodies — what production runs — records neither. Method and destructor call edges reach invalidation through the query-owned channel instead, as owner-qualified StableFreeFunctionDependency edges built from each reached body transaction's references. This is the same shape RUE-1204 removed for free functions.

Which channels are live

Which phase records a channel is what decides whether it is vestigial, and there are three:

Phase	Status	Channels
Whole-program body analysis (analyze_function_bodies_lazy)	vestigial — production runs compose_queried_bodies	free-function (removed by RUE-1204), named-method, named-destructor
Declaration binding (sema/typeck.rs)	live — both body paths bind declarations	declaration types, named consts
CFG construction	live — both body paths build CFGs	implicit named destructors

Implicit named destructors are therefore untouched. A probe on a fixture with a method call and a drop fn, run before deleting anything, reports:

methods = 0, destructors = 0, implicit_destructors = 2, free = 5

Zero named-method and zero named-destructor edges against two implicit ones — so only the first two are removable, and the emptiness is measured rather than inferred.

What changes

• Delete both projections in the manifest translation, their *_events_translated counters, their tuple slots and completeness slots, and the StableNamedMethodDependency / StableNamedMethodDependencyTarget / StableNamedDestructorDependency types that only they constructed.
• Drop the corresponding SemanticDependencyInputManifest fields and accessors, the CanonicalSemanticOutput accessors that are now unread, and the unstable.rs views over both. The CanonicalSemanticOutput fields themselves stay: they still feed unstable_parity_snapshot, so that snapshot is unchanged.
• Update the differential oracle's fingerprint format string, which formatted both removed fields.

SemanticDependencySurface::GenericNamedMethodCall stays. Its manifest producer was vestigial, but a second producer is live and independent of the projection: every generic body owner takes a per-owner blocker on that surface. NonGenericNamedMethodCall and NamedDestructorCall lose their only producers and go, along with the now-unreachable DestructorEndpointUnavailable reason.

Behavior

The invalidation plan folded these edges into direct_dependencies and into the reverse-dependency closure. Both folds ran over permanently empty vectors, so removal is behavior-neutral. Confirmed by test rather than by inspection: incremental_invalidation_closes_across_method_and_destructor_bodies edits a leaf called from a method body and asserts the plan still closes over the calling method and its caller (invalidated = ["leaf", "main", "run"]) while an unrelated function stays reusable.

No public surface outside the compiler consumes any of the removed names, and unstable.rs is explicitly outside the supported facade's compatibility policy.

Verification

Each run captured by exit code:

• scripts/rue quick — pass (debug-assertion policy valid, 92 reviewed; no debug_assert! added or removed, so the per-file counts in scripts/validate-debug-assert-policy.py are unchanged)
• ./clippy.sh — pass
• scripts/rue premerge — pass (78 targets, 0 failures)