test(async-replication): cover the events handler (7/8)

[marceloneppel]

Issue

Covers the events handler ported in 3/8 and 4/8, driving real framework events so the guards, defers and substrate branches stay in the path.

Solution

• create-replication: already-set-up guard, success path (counter + relation name + status), stale-counter clearing before the guard (DPE-10203), mirrored-counter kept, consumer-side rejection;
• promote_to_primary: no-primary failure, read-only standby promotion with the per-substrate status message, stale-counter clearing;
• relation joined/created: unit-address publication and counter propagation;
• relation changed: late-joiner early exit per substrate, leader stop + defer until all units stopped, VM relation stop-marker;
• relation broken: standby vs primary counter handling, ModelError-resilient standby check, departing-unit early exit;
• secret_changed consumer credential sync, set_app_status messages, handle_read_only_mode, and the _same_secret_id helper.

Known coverage gaps against the charm originals (flagged by the pre-merge faithfulness review, follow-up candidates): the K8s standby-pgdata-cleared once-only gating, the force-promotion reachability failure path, the non-leader _wait_for_standby_leader paths (isolated restart / defer), _configure_primary_cluster's primary-side branches, and the offer-side secret_changed branch — the latter three are additionally exercised by the adoption PRs' adapted charm tests.