Flaky-alert scheduling no longer skipped for processor-ingested test runs

Runs ingested through the processor role now schedule flaky-alert evaluation correctly. The processor lacked read access on automation_alerts, so Repo.all(Alert) inside enqueue_flaky_alert_evaluations raised a 42501 error in the async task. The run still ingested fully, but flaky-alert scheduling was silently skipped. The grant set now covers the full Postgres ingestion surface (accounts, projects, webhook_endpoints, automation_alerts, oban_jobs, oban_peers), applied atomically on next deploy.