After 4 surgical fixes the state_messages_published_on_snapshot_broadcast
test still reports 'expected ON state, got []' on CI — and we can't
tell whether the publisher is publishing nothing, or publishing the
wrong topic, or publishing to a session the subscriber lost.
Two changes to surface what's actually happening:
1. Widen subscription from `homeassistant/binary_sensor/+/presence/state`
to `homeassistant/#`. Now the captured-message dump shows every
topic the publisher emitted under the homeassistant prefix —
discovery configs, availability heartbeats, state messages,
anything else. A narrow filter was hiding which side of the
pipeline was broken.
2. Add stderr `[diag]` lines that dump every captured (retain, topic,
payload-prefix) on test failure. CI runs `--nocapture` so the lines
land in the workflow log. From the next failed-CI log we'll know
whether:
- publisher isn't emitting state at all (no /state topics in dump)
- publisher is emitting to a different topic shape (typo in
topic format string)
- subscriber connected to a stale session and missed messages
(would see discovery + no state but dump would have count > 0)
- subscriber is connecting after publisher disconnected (count = 0
even after widening)
This is a debugging commit, not a production fix — once we know the
exact failure mode from the next CI log we can ship a real fix.
Refs PR #778, issue #776.
Co-Authored-By: claude-flow <ruv@ruv.net>
636ca7b52f