Dead letter dict structure to reliably re-inject failed KytosEvent on demand

[viniarck]

This is for adding support for a dead letter dict (data structure) to hold failed KytosEvent that can happen for certain cases when handling asynchronous where returning an error for a caller isn't applicable as mentioned on the no-sql DB blueprint PR #188. Also, there should be a way to re-inject them on-demand as needed through an internal endpoint, so since kytos implements a simplified message bus of routed events, ultimately, this will be a safety net when a full outage of the DB happens (or any other unhandled exceptions, but the primary driver is the DB failures) in an worst case scenario, which is expected to be used rarely.

The DB cluster is supposed to provide high-availablity, initially we'll have 3 nodes, and we can keep increasing and adapting as needed to minimize a full outage likelihood even more.