Summary

Salvage of #33263 by @donovan-yohan onto current main.

Changes the gateway-embedded Kanban dispatcher's corrupt-board cache from indefinite suppression to a 5-minute quarantine. After the quarantine expires, the dispatcher retries the same board even when the DB file fingerprint hasn't changed, while still avoiding per-tick retry/log/backup spam. Also catches KanbanDbCorruptError (a RuntimeError) in addition to the raw sqlite3.DatabaseError corrupt-board path.

This is a recovery-layer mitigation, scoped to making the gateway un-stick itself instead of staying wedged on a single board fingerprint until restart. It is not a fix for whatever is actually corrupting the DB.

Closes #33263. Related: #32543, #32593.

Changes

• gateway/run.py: disabled_corrupt_boards now stores (fingerprint, monotonic_at) instead of just the fingerprint; tick retries after CORRUPT_BOARD_RETRY_AFTER_SECONDS = 300 elapsed even on unchanged fingerprint. _is_corrupt_board_db_error now recognises KanbanDbCorruptError. Added a separate except Exception branch that re-routes corrupt-guard raises through the same quarantine path.
• tests/hermes_cli/test_kanban_core_functionality.py: parametrised the existing disable-without-traceback test over sqlite3.DatabaseError and KanbanDbCorruptError. Added test_gateway_dispatcher_retries_corrupt_board_after_quarantine.
• scripts/release.py: AUTHOR_MAP entry for @donovan-yohan.

Validation

• scripts/run_tests.sh tests/hermes_cli/test_kanban_core_functionality.py — 167/167 passing.
• Targeted: pytest tests/hermes_cli/test_kanban_core_functionality.py -k "gateway_dispatcher_disables_corrupt_board_without_traceback or gateway_dispatcher_retries_corrupt_board_after_quarantine" — 3 passing (parametrize × 2 + new test).

Note on the wider cluster

This PR addresses the recovery dimension only. Open separately:

• Backup-file churn (_backup_corrupt_db is unconditional on every _guard_existing_db_is_healthy failure): fix: suppress repeated corrupt kanban backups #33319, fix: harden kanban corrupt board quarantine #32094.
• Probe false-positives under concurrent WAL writers: fix(kanban-db): retry integrity probe before flagging DB as corrupt #31795 (retry probe), fix(kanban): remove false-positive corruption detection from separate probe connection #32449 (remove duplicate probe).
• Actual corruption mechanism (this report's DBs show intact SQLite header but malformed internal pages — does not match the Interrupted OpenAI/httpx request thread survives across turns and writes TLS record bytes to unrelated file descriptors on delayed close #29507 TLS-FD-recycle signature): [Bug]: Kanban DB intermittent corruption after worker crash — missing WAL checkpoint #32543, [Bug] Kanban DB corruption when gateway and dashboard open the same board DB concurrently (WAL mode) #33169, feat(kanban): add busy_timeout PRAGMA to prevent WAL corruption under concurrent writers #32532, [Bug] kanban.db corrupted on WSL2 with multiple gateway processes #32749.