Introduce explicit adversaries

Ideally we would test the protocols with explicit adversaries. This is a huge topic in its own right however. There _is_ some work by the researchers on defining "ideal" adversaries; if this research comes to something we can use, that would be a good starting point for this ticket. Otherwise, this might be too ambitious for now. Bear in mind that the network problems that we _do_ extensively test are _already_ a kind of adversary from the point of the crypto, so we already cover this to _some_ extent.

One thing we should check at least is that if a node produces an invalid block, the other nodes properly disconnect from it.