feat: capture the whole wire surface, not just declarations

[cayossarian]

golden_tree.json holds $description documents, because conformance asks what a device declares. That cannot exercise a consumer: a parser fed only descriptions can be asked whether it understands the shape of a panel, never whether it builds the right snapshot from one — which is the part that reaches a user.

golden_wire.json holds the retained surface instead: descriptions, $state, and every property value, keyed as a consumer receives them.

37 devices, 37 described, 494 property values

span-panel-api-schema-1 vendors it and drives its parser end to end from it. That is why it is published as an artifact rather than kept internal.

The capture goes through production wiring

start_clone grows an optional publisher parameter, and the capture uses it. Reassembling the emitter inside the script would make it a second copy of that wiring, free to drift — and a capture taken through different wiring than a real panel uses proves less than it appears to.

MqttPublisher was already a Protocol and CloneRuntime.mqtt already held that type, so this is the existing seam being used rather than a new one. No broker is needed.

Shape is asserted; values are not

noise_factor perturbs power and current and the clock advances, so values differ every run. The test compares which devices and which topics — a property added, removed or renamed fails it, a different wattage does not.

That is the opposite policy to golden_report.json, deliberately: a conformance profile must not drift silently, while a fixture of noisy floats held to byte equality produces only failures nobody can act on.

Mutation-checked — injecting one phantom topic into the committed fixture fails the staleness guard by device id.

Placement

test_boundary.py caught my first attempt, correctly. The capture lives in emitter_adapter, not conformance: taking it means running this simulator, while the conformance rules describe any eBus publisher's output. The test sits in tests/conformance/ because the fixture it guards does.

Verified

Check	Result
Test suite	432 passed (5 new)
pre-commit run --all-files	all hooks pass
mypy --strict on the changed runtime	clean
Staleness guard	fails on an injected topic, passes on restore
Wire and tree captures	same 37 devices

Known gap, unchanged by this PR

grid/islanding-state is still absent, because no config can produce a MID. The emitter supports one fully — wire/profiles/mid.json declares the grid node, bag_builder resolves ("mid", "grid/islanding-state"), snapshot carries islanding_state, and Emitter iterates of_class("mid") — but spec_generator.build_manifest() never creates a MID DeviceInstance. So the gap is one missing instance in the manifest builder, not a config or capability gap.

It is the single property span-panel-api-schema-1 reads that this simulator does not exercise, and it is the one whose mapping was corrected most recently (grid_state reads islanding-state, not grid-state). Worth closing, but it changes what the simulator publishes and would require recapturing both fixtures, so it belongs in its own PR.