If auditNetworkProducers is off (as it is by default) a reforward (because of a missing send reply) will be a duplicate.
If the initial messages is still in the store the duplicate will be trapped by the message cursor but the message remains in the store. So on a restart the cursor is fresh (unless in the case of jdbc when there are still messages in the store) and the message gets redispatched. In the jdbc case - if the store/destination is empty there will be a resend b/c there is no sequence state to replay.
Duplicates detected by the cursor need to deal with the duplicate message - moving to the DLQ makes the most sense - the cause can indicate the reason to allow separation from poison due to redelivery failure.
Client duplicate detection currently uses a poison ack, so this makes the behaviour consistent.