Add a way for NetworkBehaviours to signal an error to the swarm.

[twittner]

A NetworkBehaviour may enter an error state and can not proceed in an orderly way. The current API contract provides no means to signal such a condition to the swarm which could then act appropriately. Instead of panic! we may want to support more fine-grained ways to handle this, e.g. reset the state of the behaviour or terminate the swarm. One possibility would be to add NetworkBehaviourAction constructors, e.g. ResetBehaviour and TerminateSwarm. When receiving a ResetBehaviour the swarm would provide the behaviour with the current peer information so it could clear its internal state and assume the given peers as connected. On TerminateSwarm the swarm would itself shutdown operation.

[tomaka]

As for TerminateSwarm, the only way the Swarm could handle that is to return an error to the user, and the user would then have to destroy the Swarm. In other words, Swarm producing an error would mean that TerminateSwarm has been called.

To me, this mechanism would be redundant with NetworkBehaviourAction::GenerateEvent being propagated to the user. The only difference would be that TerminateSwarm would have a bit more semantics.

[tomaka]

As for ResetBehaviour, I guess it makes a bit more sense. Ideally I think that we should pass the list of nodes we're connected to by parameter whenever any of the methods of NetworkBehaviour is invoked, but that's a change that is way more complicated than it sounds.

However note that some implementations of NetworkBehaviour have to stay in sync with some state within their associated ProtocolsHandlers, and that state would be lost anyway in case of a "reset behaviour".