Task: Bring ores.nats to the clean standard

ores.nats passes the Component Clean Standard for a component of kind All.

• Every checklist item that applies to a component of kind All passes, with its evidence recorded on the task.
• Every item that does not apply is recorded as not applicable, with the reason.
• Every generator defect found is fixed at its source, with a regression test.
• ores.nats is listed in COMPONENTS_UNDER_TEST, and check_component_drift.py --all passes.
• V06 states why the first half of that criterion cannot hold: the registry admits only model-bearing catalogue entries. The experiment that listed nats anyway is kept as evidence, and check_component_drift.py --all passes without the entry.

Work the Component Clean Standard in its section order, one entity at a time, and record each item's result in the table below.

The standard's table holds 49 items, though its own prose still says 47. This component is of kind All, and it has no model and no protocol. The All items are the work. The Entity items are recorded as not applicable, and so are the Protocol items, except B04, which is worked because the component sends one subject.

1. Baseline and inventory (B01 to B06). Record the baseline before any edit.
2. Models (M01 to M10).
3. Protocol (P01 to P06).
4. Generation (G01 to G08).
5. Wiring and data (W01 to W04).
6. Shell (S01, S02).
7. Structure and hygiene (H01 to H05).
8. Verification (V01 to V08), ending with the registry entry.

Every item is now recorded.

Manual QA scenarios (scaffolded via compass add test_scenario) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

ores.nats is at the Component Clean Standard for a component of kind All. All 49 items are recorded on this task: 23 pass and 26 are not applicable, each with its reason. The Entity items do not apply, because the component has no model. The Protocol items do not apply either, except B04, which is worked because the component sends one subject. V06 does not apply because the registry admits only model-bearing catalogue entries, which is the standard's own recorded finding.

The work itself came to four things. The dead public surface the B06 census found is deleted: the JetStream read and message API, the three domain types behind it, two client verbs and the free as_string_view helper, 582 lines in all, and the four services that take an admin handle still build against the one method they call. The namespace documentation header stays, and its stale claims are corrected in place: it named a retired client and a component that no longer exists. The diagram went from an automated skeleton to a diagram with an API, relationships and the free-function facets, and the generator defect that had silently dropped two classes is fixed at its source with five regression cases. The overview now states the real wire format: resolved at startup, msgpack by default, not JSON per payload.

Every gate is green for this component: the drift gate passes with the component's own source lists up to date, and model drift, protocol twin coverage, handler permissions, populate references, the recipe inventory and the documentation checks all pass. The whole tree builds with every consumer, and the component's suite passes at 55 assertions in 28 cases. The one gate that is not green repo-wide is the CMake source-list check, which reports 97 stale lists in other components; none of them is this one, and the clean-compute task recorded that sweep as a codegen-wide finding. The physical-space inventory check reports one stale table on main, ores.doc.knowledge.org, which this branch does not touch.

What stays open is recorded rather than claimed: nine sources have no test and are the coverage work item, the session and workspace headers have no producer, and the JWKS exchange bypasses the wire codec. Notes record each.