sink/periodic-jobs: deliver and report one item at a time (DI-3015)

[jbndbc]

Migrate the sink from MessageConsumerAdapter to SinkMessageConsumerAdapter. job-store now dispatches one JMS message per item, and the sink implements deliverItem(ConsumedMessage, ChunkItem) instead of assembling a DELIVERED chunk. Header reading, result reporting and the tracking-id scope move to the framework.

The sink aggregates a whole job before delivering anything, so it overrides usesDeliveryWatermark() to false. It still reports every item individually, since the DELIVERING counters and the per-job gate are driven by those reports. The job termination item arrives as an ordinary item message carrying ChunkItem.Type.JOB_END.

Drop the five second sleep that preceded finalization. It guessed at a window the chunk protocol opened by reporting the result before committing the sink's own transaction. deliverItem commits first and the framework reports afterwards, so a reported item is an item whose datablocks are durable and the termination chunk is released only against complete data.

InvalidMessageException from finalization is now reported as a FAILED termination item rather than discarding the message, which completes the job and sets its fatal error flag instead of leaving it forever incomplete.

[ja7]

Hvordan i forhold til
it rolls the JMS session back and the item is redelivered until the broker gives up
Jeg er ikke sikker på hvordan vi håndtere det i dag. ( er det JobStore der arbonere på DeadMessageQueue ? )

[jbndbc]

Hvordan i forhold til it rolls the JMS session back and the item is redelivered until the broker gives up Jeg er ikke sikker på hvordan vi håndtere det i dag. ( er det JobStore der arbonere på DeadMessageQueue ? )

Ja, det er samme mekanisme som i det nuværende, bare lavet om til at fungere på item niveau i stedet for chunk niveau. Den funktionalitet lå i et tidligere PR.