Improve RabbitMQ Implementation

[ByteAlex]

The current RabbitMQ implementation has a few flaws.

1. Add channel pooling: Currently for each call to RabbitMQ#send* methods a new channel is created
2. Connection Pooling (?) - What's the behaviour for the underlying connections of the channels?
3. Push messages to exchanges and not directly to the queues and let RabbitMQ do it's work
4. Add incremental backoff on error in the pipeline for send/receive