feat: publish a device tree under any Homie 5 domain

[dcj]

Closes #61.

Device gains homie_domain=, defaulting to ebus. Ten publisher-side topic constructions now resolve it instead of the module constant.

What the domain now covers

• property value topics, and /set subscription topics
• $state and $description
• the retraction topics delete() and delete_all_from_mqtt() clear
• will() and the LWT installed on an SDK-owned client, which stay identical to each other by construction

Verified by listing every topic construction in homie.py: the nine Device / Property sites all resolve the domain now, and the ones left on the constant are Controller (already configurable) and the homie version field, which is not a topic.

The inbound /set check

Property._settable_callback compared the received topic's domain against EBUS_HOMIE_DOMAIN. Left alone, a device published under homie would have subscribed to the correct /set topic and then silently rejected every command that arrived, which is a worse failure than not supporting the domain at all. It now compares against the tree's own domain, with a test in each direction: accepted on the tree's domain, ignored on a foreign one.

It belongs to the tree

Like the MQTT connection and the QoS. A child under a different domain would sit outside its own root's subtree, and the root's Last Will (a single retained publish on the root's $state) could not cover it.

So only a root stores it, descendants read it through the new Device.homie_domain(), and a child passing its own gets a ValueError, matching the existing parent= plus mqtt_cfg= rule. It is refused even when the value would have matched (test_a_child_of_a_default_domain_root_is_also_refused), because the rule is structural rather than a value check, and a silently-dropped domain would surface as topics on the wrong prefix rather than as an error.

Compatibility

The default is unchanged, EBUS_HOMIE_DOMAIN stays exported, and a publisher that never mentions the parameter is byte-identical on the wire. test_default_domain_is_ebus and test_explicit_ebus_is_the_same_as_omitting_it pin that.

The Device docstring's homie_domains config for future use, not currently supported by this code stub is replaced with what to actually do.

One test-double fix

_make_wired_property in test_homie_device.py builds a MagicMock device, which returned a MagicMock from homie_domain() and broke three settable-callback tests. The double now answers it. Fixing the double rather than making production code defensive about mocks.

Out of scope

Multi-domain on one Controller. Today one Controller monitors one domain, so watching both trees means two of them. Noted in #61 as a separate question.

Verification

• 653 tests pass (636 before, 17 new)
• ruff check and ruff format --check clean
• markdownlint clean on CHANGELOG.md and README.md
• README gains a "Publishing under a different Homie domain" section beside Device Trees, since the domain is per-tree

🤖 Generated with Claude Code