r202639: measure the reference captures with the adapter fidelity harness

[cayossarian]

r202639: measure the reference captures with the adapter fidelity harness.

Each vendored reference capture becomes a fidelity cell, using the harness and rules the MAIN 32 cells use: no property is silently dropped, no value is fabricated or breaks its documented transform, and every valued drop is a reviewed row. What the cells pin is keyed by row, never by capture.

• A circuit-typed device with no breaker space has its own role (meter-outside-panel), read import-positive with imported energy as consumed, as circuits.build_circuit reads a meter outside the panel.
• The EVSE charge-current pair gets transforms in both spellings charge_limit.resolve_charge_limit reads.
• With no MID, dominant_power_source GRID is a derivation citing the elimination in panel.resolve_dominant_power_source.
• info/tags, info/locations, info/dedicated, connection/backed-up, the upstream lugs' service-rating and the downstream lugs' feeds-role are pinned as dropped by the identity and topology node rule. The site properties are pinned as dropped.

Follow-up: four str fields hold "UNKNOWN" where the panel publishes nothing for them, recorded in the cells as explicit exceptions (a pre-existing str sentinel; the type question is deferred past 3.8.0b1):

• circuit.relay_state, circuit.relay_requester and circuit.priority on a device that publishes no relay or shed priority;
• panel.main_relay_state while status/relay is unpublished.

Checks:

• pytest tests -q: 1848 passed, 7 skipped, 9 xfailed (49 new; tests/adapter_fidelity 79 passed, about two minutes).
• pre-commit run --all-files: all hooks pass.