[miki3421]

When CouchDB or the Controller stalls, cron-init can leave Jobs active indefinitely while new scans keep starting. This change bounds each Job and its HTTP/pagination work, reports failures through the process exit code, and rejects incomplete scans before dispatching actions.

Related to apache/openserverless#280.

• Keep concurrencyPolicy: Allow; add a 120-second Job deadline, resource requests/limits, restartPolicy: Never and backoffLimit: 0.
• Bound HTTP waits and scan time, pages, documents and response sizes; reject missing, repeated or cyclic pagination bookmarks. Strict CouchDB behavior is opt-in for the scheduler.
• Preserve per-action error isolation and report invocation/autoexec update failures. Do not automatically retry invocation POSTs or replay the whole batch.

This is a containment fix. It does not prevent duplicate invocations, recover missed executions, or provide exactly-once delivery. Allow can still overlap scans. The current time-window and autoexec semantics remain best effort; durable catch-up and idempotent acceptance require separate work. Read timeouts bound inactivity; the Job deadline is the final bound for a slow trickling response. Existing Jobs do not inherit the new settings.

Validation: 102 local tests and executor doctests passed; 27 focused tests per candidate image passed. Baseline and ADV each passed 13 real-process HTTP fault/control scenarios, Kubernetes deadlines for Running and unschedulable Pending Pods, and Allow-overlap checks. Real synthetic Alpha/Beta actions covered multi-page scans, periodic execution, autoexec and not-due actions; each environment also passed 15 application checks with zero load errors. Temporary test resources were removed. Laboratory overlays preserved their existing CouchDB authentication and ADV-specific helpers.