Compile door: do not refuse module shells as member_not_a_binder

[gunbai-bot]

Summary

• Verdict is (b): a production compile-door defect, not a single-root harness artifact. compile_source_root_ingest_with_admission assembles a trivial module, then the subtree roster (unit_modeling / lifecycle_carrier) treats every TypeNode as a record and refuses Authored module-shell members as member_not_a_binder. The door ledger already measured that fatal on dag/std/error_primitives.dag and dag/std/bit.dag.
• v2.cli.compile_cli does not call this door (emit_closure_from_ingest_located). Callers that do: compile_door_ledger, compile_ingest_with_admission / compile_dag_source_to_target_text, stage_translate.
• Fix: shared carrier predicate v2.lens.common.record_field_carrier skips module shells, marked bodies, and graft spines. fact_density treats graft identity atoms as NotATypeCarrier so the skip is not unmasked as hollow-alias. Hand-built name→type fields still refuse. Door-grain mandatory_tag controls (mandatory_tag gate: read the lowered data-declaration carrier (clean fixture refused at normalized grain) #13513) are now authorable.

Test plan

• Floor claims v2.test.lens_unit_modeling.module_container_is_not_a_record_carrier (GREEN grafted shell/spine; RED hand-built field still member_not_a_binder on the subtree roster)
• Floor claim v2.test.lens_fact_density.surface_identity_atom_not_a_type_carrier
• Existing binder-node RED a_lens_refuses_a_field_that_is_not_a_binder still holds
• After merge, door-grain mandatory_tag RED/GREEN through compile_source_root_ingest_with_admission can name this lens's reasons; retire the member_not_a_binder ownership row when the long door-ledger count is zero

Made with Cursor

[gunbai-bot]

Parent asked not to open a PR; this item is folded into integration/bold-bee-114 (#13560). Branch kept for the closeout report.