rung_drop infer facts first-wins: route back is site keying

[gunbai-bot]

Updates the declared rung drop gunbc.rung_drop.infer_facts_key_conflict_reverted_to_first_wins (landed with the #12852 revert) to the facts-key ruling of 2026-10-01 (neat-boar-16, option B).

• Replacement: the list-spine-end-label route is withdrawn because it covered only list terminators. The route back is now infer's facts keyed by SITE: the injective (edge label, ordinal) path from the root of one InferredTree, declared as a keying relation at the DESIGN §3b keys home, with every facts lookup converted and the infer_facts_key_conflict refusal re-landed in the same change.
• Population: unchanged in membership. The two childless-Conj claims I first added were removed after calm-hawk-793 showed by execution and by structure that they are red from an inhabitants-roster-membership rule, not the facts key (their trees have no equal key partner). The measured specimen stays kc_g/kc_f.
• Restoration trigger now names the capability (DESIGN §4b(3)): an injective site key, no reader left on the structural Node key, a key unusable outside its own tree, the refusal restored on the real path, and kc_g/kc_f inferring with each terminator carrying its own fact.

The docs/design-rung-drops.md projection was regenerated by generated_artifact_gate main_wet_one at the pushed head (rc=0).

No code changes; the site-keyed PR follows.

🤖 Generated with Claude Code