Namespace scaffold: reference-derived module-graph edge producer (flagged; no grammar delete; module_graph.dag only)

[briansrls]

Summary

Introduces the namespace migration scaffold in v2.lens.module_graph so module-graph dependency edges are edge-source-agnostic before the import-grammar delete lands.

• Adds ModuleDependencyEdge as the unified edge row type and routes closure/touched-path queries through dependency_resolution_facts_live / dependency_closure (today still import-backed).
• Adds reference_derived_resolution_facts_flagged with an explicit Scaffold disposition — the future swap target for parsed-tree / container.member reference edges. Not wired into the live producer yet; interim body maps import facts through ModuleDependencyEdge (byte-identical on the import-only corpus per reference_closure_b2_safety_wall_holds).
• No grammar delete; module_graph.dag only.

Test plan

• cargo run --bin gunbc -- run --source-root dag --source-root src/v2 --entry dag/test/claim/long/import_closure_live_test.dag --function dag_import_closure_live_witness_bundle_holds --claim-run — green
• cargo run --bin gunbc -- run --source-root dag --source-root src/v2 --entry dag/test/claim/long/reference_closure_equivalence_test.dag --function reference_closure_b2_safety_wall_holds --claim-run — green
• CI floor (dag_compile_clean_gate + witness batches) via PR checks

[gunbai-bot]

Addressed the claude-opus naming finding in 395a6a065: renamed the scaffold producer to reference_derived_resolution_facts_flagged_pending_reference_producer and updated the dissolution note to state explicitly that the body is still import-backed (interim stand-in), not parsed-tree reference-derived yet. The Scaffold disposition bind tracks the new name.

The earlier CI failures were from a superseded run where emit_determinism was cancelled mid-flight when a new push landed; the prior ci job on dd1691f completed green. Re-running on 395a6a065 now.

— sent from sharp-stag-26