fix(dev-setup): install ebus-emitter from its lock; bump to 1.0.12

[cayossarian]

Problem

Starting the simulator failed for every panel:

[homie] ERROR: reason=deviceConnectBrokerFailed,id=sim-40t-001
  File "ebus_sdk/homie.py", line 1955, in connect_broker
    self.mqttc = MqttClient.from_config(mqtt_cfg=self._mqtt_cfg, ...)
  File "ebus_mqtt_client/client.py", line 240, in from_config
    endpoint = mqtt_cfg.get("host", MQTT_DEFAULT_HOST)
AttributeError: 'NoneType' object has no attribute 'get'

Root cause

scripts/dev-setup.sh installed the emitter with a bare uv pip install --editable <path>. That re-resolves the emitter's dependency constraints against PyPI and ignores the emitter's own uv.lock (which pins ebus-sdk 0.1.5). A fresh bootstrap therefore resolved ebus-sdk>=0.1.5 to 0.12.0.

Per the SDK's own metadata: "0.2.0 introduces parent/child device trees and contains breaking changes to the Device constructor." Among them, mqtt_cfg's default changed from {} to None. The emitter's build_graph() constructs the root Device as a passive topic/schema model with no mqtt_cfg, so connect_broker() dereferenced None before the graph was ever built.

Minimal repro, no simulator involved:

ebus-sdk 0.12.0: ebus_sdk.Device('r', name='R', type='ebus.panel') -> AttributeError
ebus-sdk 0.1.5:  -> OK

Fix

• dev-setup.sh now installs the emitter's locked runtime deps (uv export from its lock), then the emitter itself with --no-deps so nothing is re-resolved. The venv matches exactly what the emitter pins.
• ebus-emitter carries a matching ebus-sdk>=0.1.5,<0.2 upper bound (committed separately in that repo), so the constraint is enforced declaratively too.
• Version bumped to 1.0.12 to match this release branch; sync-version.sh propagated it to __init__.py, config.yaml, and the Dockerfile label.
• Root CHANGELOG.md [unreleased] stamped as 1.0.12, and the add-on CHANGELOG.md — which had stopped at 1.0.11 — gets a 1.0.12 section covering the emitter cutover and the schema changes downstream consumers will notice.

Verification

• tests/test_panel.py::TestPanelInstance::test_start_and_stop was failing with the identical AttributeError. It already covered this; it just had not been re-run after the venv drifted. Now passes.
• Full suite: 227 passed. Emitter suite in its own venv: 150 passed.
• Started the app end to end: Panel sim-40t-001 started (config=MAIN_40.yaml) / Reload complete: started=1, stopped=0, reloaded=0, errors=0.

Follow-up, not addressed here

The emitter uses ebus_sdk.Device purely as a topic/schema model — Publisher sends everything over the producer's aiomqtt client and only reads get_device_id()/get_node_id()/id(). But on 0.1.5, constructing that model also opens a second, unmanaged paho connection to the default broker and publishes $description/$state on it. That is pre-existing, not a regression, and it is the same knot that makes moving to ebus-sdk 0.2+ non-trivial: the newer SDK requires a root Device to own a broker connection, which conflicts with the emitter's "producer owns the connection" design. Worth settling before raising that version bound.