[DP-104] spawn without hearbeats

[Imported from JIRA. Reported by Facundo Dominguez @facundominguez) as DP-104 on 2015-02-24 11:57:14]
by Facundo Dominguez at https://cloud-haskell.atlassian.net/browse/DP-104:

It looks like n-t-tcp does not implement heartbeats, nor does the n-t documentation ask for that. Yet, spawn seems to rely on monitoring to learn of disconnections.

spawn uses delay, which won't run the spawned process unless it gets either a monitor notification or an ack from the caller. On the other hand, the caller can complete the call to spawn before the ack to delay arrives. If a disconnection occurs at this point, the spawned process will block indefinitely.

A simple fix is to stop using delay, and ask the user to take extra precautions if the spawned action is killing the remote node.

A simple workaround is to have the user use spawnAsync directly.

If call starts using cpDelayed as proposed in https://github.com/haskell-distributed/distributed-process/pull/183, the spawned process could block indefinitely if there is a disconnection. The call to call would not complete, but there wouldn't be a way to interrupt it while avoiding leaking the spawned process.

A fix for this could be to give up monitoring the spawned process and avoid using cpDelayed.

Thoughts?

Start by tracing spawn, delay, spawnAsync, and call, then review how n-tcp monitoring and heartbeats handle disconnections. Compare the cpDelayed proposal in pull request 183 and determine the intended behavior when disconnection occurs before the acknowledgement; done should prevent the spawned process or call from blocking indefinitely without leaking work.