Investigate if the Twins framework can be used for BFT-consensus testing

[dnkolegov-ar]

The Twins methodology tests Byzantine behaviour by simply instantiating “twin” copies of correct replicas and using network partitioning, giving a lightweight but powerful fault model. However, its assumptions and APIs may not align with the current architecture.

The reference implementations:

• era-consensus
• relab

[patrick-ogrady]

As discussed, there are a few layers of defense against both connecting to duplicate peers and handling messages from them. I think the cleanest way to implement this would be to "share" a simulated p2p channel between multiple logical consensus instances. We'd need to create a wrapper around "receive" to duplicate messages that the peer is supposed to receive but wouldn't be too hard at all.