Provide MxAgents with a dual state model

It's often the case that whilst initialising an agent might need to wait on other actors, so we should either copy the init >>= serve model from client-server, or provide a dual state monadic environment so we don't have code like this...

data AgentState = Booting | ActualState { f1 = ..., f2 = ... }

-- and inside a listener
startIt = mxAgent agentId Booting [ mxSink handlePeersReady, mxSink handleEvent ]
where
  handleEvent Booting = mxSkip 
  handleEvent ActualState{...} = ...

And worse still, have to worry about the boot state when you're buried in implementation code...

forwardToServer _   HBAgentBoot = liftMX terminate  -- state invariant
forwardToServer ev' HBAgent{ hbServer = sPid }
                                = liftMX (send sPid ev') >> mxReady

handleRestarts _   HBAgentBoot = liftMX terminate   -- state invariant
handleRestarts pid sta@HBAgent{ hbServer = sPid' }
        | pid == sPid' = restartBoth sta
        | otherwise    = mxReady

On the other hand, we want this API to remain as simple as possible, since here we're providing service and utility, and policy belongs elsewhere.