Absence — why a factor is NOT in the corpus

Open (no key). The reasoned counterpart to /v1/coverage: where that reports a zero, this reports why. Each record carries the subject, the publisher we checked, the route we tried, a confidence and a review date. Five classifications: "structural" (no publisher issues this anywhere — stop retrying), "not_yet_sourced" (a publisher exists and is named; our backlog), "refused" (we found it and declined; the reason is stated), "held_not_counted" (we DO hold it — see we_hold for the key — and the coverage metric cannot see it), and "coupled" (zero only because another family is zero). These are authored judgements about what the world publishes, not values derived from our corpus, so each carries a review date and reports stale:true once past it.