Checkpointing and restarting from checkpoints

[jjallaire]

For long running samples we may want to save the event stream as a checkpoint, and allow agents/solvers to re-constitute themselves from a checkpoint.

Note that this can't be done in completely general way where any solver can be checkpointed (as there could be in-memory state, etc.) however a cooperative scheme is possible whereby solvers can be offered a checkpoint and restore themselves from it if they can. It seems like basic_agent() and other more complex agents could be made to work with the scheme (although sandbox state is another confounder here -- if we assume the sandbox survives after the task exits this could be made to work).