kafka: retire unhosted logs when bridge metadata deletes a topic

[bringhurst]

Retire an unhosted local log when metadata reports its deletion, behind the cleanup flag. Preserve hosted replicas, unrelated logs and stale-controller fencing. Point process CI at the companion 3.0 repair.

Verification and release boundary

The latest clean-source mixed migration and process-evidence audit pass locally. The current wrapper suite passes 132 tests, with unchanged Kafka dependencies before and after the run. The reorganized Python suite passes 65 tests. These results apply to the complete candidate stack, not every intermediate commit. Focused tests are included with the relevant layers.

The final requirement audit and remote CI review remain open. Published artifact qualification, live production state, the client/tool floor, the deployed ZooKeeper server, capacity limits and named security/release approvals remain release gates. Do not deploy an intermediate stack commit.

[bringhurst]

Blocking qualification finding: the current passing migration scenario does not cover a former replica that stays offline through reassignment, deletion, and name reuse. A targeted run reached ISR and promoted that replica, but read the old 64-byte record at offset 0 where the recreated topic had a new 128-byte record. The old broker missed the deletion notification. This also affects the companion in #577. I am extending the cleanup gate to reconcile unhosted logs from the complete initial metadata image, preserving assigned leaderless logs and same-epoch incremental behavior, and adding the case to mandatory qualification. Do not treat the existing green checks as approval to deploy.