fix(clone): read circuit energy seeds in the enclosure reference frame

[cayossarian]

Companion to ebus-emitter 0.2.1, which corrected the same inversion on the publish side.

The bug

clone.py seeded initial_consumed_energy_wh from a scraped panel's imported-energy and initial_produced_energy_wh from its exported-energy.

The wire is enclosure-framed: exported-energy is energy the enclosure exported to a circuit (normal load consumption), imported-energy is energy it imported from a circuit (backfeed). So both were read backwards.

Why it went unnoticed

The two repos were wrong in a mutually cancelling way — clone read imported-energy into "consumed", and the emitter published "consumed" back out as imported-energy. A cloned panel therefore round-tripped its wire values faithfully while every value carried the wrong meaning. Nothing looked broken from inside the simulator.

It only became visible from outside: the SPAN HA integration, which reads the enclosure frame correctly, displayed every load circuit as producing energy and consuming none.

Fixing only one side breaks the round-trip, which is why these two changes move together.

What changed

• clone.py — initial_consumed_energy_wh now seeds from exported-energy and initial_produced_energy_wh from imported-energy, in both the initial-translation path (_translate_circuit) and the refresh path (update_config_from_scrape).
• config_types.py — the field comments described the old (wrong) sources.
• tests/test_clone.py — the fixtures encoded the same inversion: a load circuit had a rising imported-energy and the backfeeding solar circuit a rising exported-energy, the reverse of what a real panel publishes. One comment also described positive active-power as "export", when on the wire it means the enclosure is importing from the circuit. Fixtures, comments, and the two energy-seeding test names now describe the enclosure frame.

Requires

ebus-emitter >= 0.2.1. Pairing this release with an older emitter reinstates the inversion.

This PR deliberately does not add a version guard for that. An earlier revision enforced a minimum emitter version in scripts/dev-setup.sh, but the emitter is being vendored into this repo as a follow-up — the fork is permanent, so an external path dependency no longer buys anything and the add-on image cannot install it at all. Vendoring removes the cross-repo lockstep problem outright rather than policing it, so the guard was dropped rather than landing days before its deletion.

Upgrade note

Existing cloned configs keep their initial_* values; those seeds were captured under the old reading, so a re-clone is needed for them to carry the right meaning.

Verification

• 241 tests pass
• ruff check clean, mypy clean (58 source files, no ignores added)
• pre-commit hooks pass, including sync-version across all four version references