Repair Memory Growth & Resource Leaks

I would first separate JavaScript heap retention from sockets/buffers and other external memory, then map each resource to one owner and one teardown path. The strongest suspects from the supplied evidence are per-job listeners that are never removed, retry timers surviving terminal job states, and payload buffers retained by closures or retry state. I would make registration idempotent where ownership is process-wide, bind job-scoped listeners/timers to a disposable lifecycle, clear retained buffers at terminal transitions, and preserve the shared HTTP client rather than creating clients per job. A deterministic harness should run many synthetic jobs with forced success/failure/retry/cancellation, record listener/timer/handle counts after each batch, force GC only in the isolated test when available, and compare post-settle heap/external-memory plateaus. I would not claim a production leak is fixed until the supplied profiler/metrics confirm the same retention path.