Third cause of the state_messages_published_on_snapshot_broadcast
failure (after timing fix in 5ed8e3451 and client_id fix in
2aeed32a7): the subscriber's eventloop was NEVER polled between
`client.subscribe(...).await` and `collect_published(eventloop, ...)`,
so the SUBSCRIBE packet was only queued in rumqttc's outbound channel
— it didn't reach the broker until collect_published began polling.
By that time the publisher had already emitted all 6 state messages.
The retained ones (binary_sensor presence with retain=true) should
have been redelivered on the late subscribe, but only the LAST one
would land — yet CI was reporting `got []` (zero messages).
Theory: the broker may not redeliver retained messages reliably when
the subscribe arrives during the publisher's burst, OR the test's
collect_published timing budget runs out before redelivery completes.
Fix: drain the subscriber's eventloop inside `subscribe_client` until
we see the SubAck for our subscribe. That guarantees the subscription
is active at the broker BEFORE the function returns, so non-retained
publishes from the publisher's send loop arrive normally.
Also made the subscriber client_id include a per-call nanosecond
suffix so subscribers in back-to-back tests can't collide on a shared
ID (paranoia, complementary to the publisher-side fix from
2aeed32a7).
Refs PR #778, issue #776.
Co-Authored-By: claude-flow <ruv@ruv.net>
967cede74d