handle message broker failures

[jonenst]

We have nothing to handle spurious redeliveries, we should implement something.

spurious redelivery because of failures at publish time:
https://www.rabbitmq.com/docs/reliability#publisher-side
"Therefore consumer applications will need to perform deduplication or handle incoming messages in an idempotent manner."

spurious redelivery because of failures at consume time:
https://www.rabbitmq.com/docs/reliability#consumer-side
"In the event of network failure (or a node failure), messages can be redelivered, and consumers must be prepared to handle deliveries they have seen in the past. It is recommended that consumer implementation is designed to be idempotent rather than to explicitly perform deduplication."

we should use all information provided by rabbitmq (e.g. redelivered counts) to limit the effects of duplicated messages, avoid extra computations if possible.