Re-home LayerPrefix classification onto v2.std.layer

[gunbai-bot]

Summary

• artifact → refinement is ESSENTIAL. GeneratedArtifact is a Refined<Artifact> produced by refine / Validation. That is inhabitance of v2.std.refinement, not a helper parked in the wrong module. No model break.
• cross_tree.resolution → layer was INCIDENTAL. All 20 qualified v2.std.layer.* uses were layer_prefix_from_qualified_name / layer_prefix_from_dotted_qualified_name. Resolution's actual job (source-root index, cross_tree_edge_decision) never reads LayerPrefix. The walk belongs with LayerPrefix in v2.std.layer (DESIGN §3). Moved there; citations (doc graph, host scaffold, plan) follow the symbol.

Dependents closures (one instrument)

Producer: cli_run::dependency_resolution_facts at #13526 head e78c3b8349, pool/importer roots dag + src/v2, no excludes. Reverse BFS over import_module edges keyed by module_declaration_facts paths; seed excluded.

The earlier 2476 on refinement was a different walk (import-line strip, not this producer). Same producer, both trees:

seed	#13526 head (before re-home)	#13526 producer + this re-home
v2.std.layer	1334 (resolution_depends_on_layer=true, layer_direct=9)	16 (resolution_depends_on_layer=false, layer_direct=8)
v2.std.refinement	1834 (artifact_depends_on_refinement=true)	1834 (unchanged; essential)

Facts: 55063 before / 55064 after.

Checks

• cargo clippy --all-targets -- -D warnings: clean (remote).
• Host scaffold: layer_prefix_from_dotted_module_scaffold_marker_is_declared passed; cli_run.rs compiles under clippy.
• gunbc compile --output-dir … --source-root dag --source-root src/v2 of src/v2/std/layer.dag, from_dotted_string_test.dag, and dag/gunbc/doc_graph_roots.dag: exit 0 (DeclarationRef now names v2.std.layer / layer_prefix_from_dotted_qualified_name).

Do not merge.

[briansrls]

APPROVE at exact head fa159d0ea97748b468be2cd3b9438e1499b81af7, against the head's DESIGN §3. No blocking findings.

The ownership repair is substantive rather than an edge-count workaround. Both classifiers now live beside the LayerPrefix coproduct they produce. Their definitions are removed from v2.std.cross_tree.resolution, with no forwarding alias or second .dag classifier retained there. The inspected resolution module's source-root and cross-tree admission logic does not need LayerPrefix; its layer coupling was confined to the moved classification functions. The new home depends on the existing qualified-name/algebra utilities rather than importing the old resolver to do the work.

The moved decision bodies are unchanged apart from qualifying LayerPrefix at its own home. Empty input, the lone-v2 case, each named prefix and both unknown-prefix fallbacks retain the previous result. The dotted entry still calls qualified_name_from_dotted_string before the same qualified-name classifier. There is no narrowing of the classifier domain or deletion of a decision to manufacture a smaller dependency closure.

The six-file diff is consistent with that scope. gunbc.doc_graph_roots repoints its DeclarationRef to the new owner; the plan and layer_import_scan/Rust scaffold citations follow the symbol. The pre-existing layer_prefix_from_dotted_module Rust mirror is NOT removed or changed semantically, and its explicit scaffold/dissolution obligation remains. Thus this is one .dag authority with the existing declared host mirror, not a claim of having eliminated that mirror.

Caller precision: from_dotted_string_test.dag is not edited by this PR. Its existing calls are bare names, as is the inspected grounding use; the unique .dag definitions retain those names at the new owner. The reported targeted compilation of layer, the dotted-name witness and doc_graph_roots corroborates their new binding, but remains author-run evidence rather than a compiler run I performed. The doc-graph change is an owner-qualified DeclarationRef repoint, not a new wrapper or import-based duplicate.

The essential artifact -> refinement relationship is left intact. The inspected artifact source actually consumes the refinement types/operations; this PR does not delete or relabel that dependency to improve the measurement.

Evidence boundary: the 1334 -> 16 reverse-dependents reduction is the author's paired measurement using #13526's producer at e78c3b8349, roots dag + src/v2, no exclusions and seed excluded. The reported refinement population remains 1834 on both sides. I have not independently rerun that producer, the reverse traversal or the targeted compiles; I do not treat these figures as a measurement on an independently executed build of this head, or as an approval of #13526 itself. GitHub code search was used for discovery, with relevant files read at this SHA, not as an exact-head whole-tree absence certificate.

CI independently confirmed: workflow 37572035985 names this exact SHA and completed successfully. Its five jobs—emit-build, floor, generated, rust-unit-tests and witnesses—succeeded, including the generated lane's all-target lint step. I am not promoting the green aggregate into separate verified per-claim execution/timing receipts. No new Rust libtest or unit-budget exception is introduced by this move.

No further code, test-lane or receipt change requested. This approval does not override the PR body's do-not-merge instruction; I have not merged it or performed any hardware action.