Take a more principled approach to simulation startup and shutdown

[RobertJacobsonCDC]

Now that negative time exists, the exact time a simulation both begins and ends may not be known at the code cite that needs that information. We have no mechanism to "run this code at simulation start time, whenever that time may be," nor to "run this code at simulation end time, whenever that time may be."

We can carefully manage our model in our client code so that we know these times, but this is fragile and a burden on the modeler. It'd be much better to have it as an ixa primitive.

Use Cases

Startup

• create the ValueChangeCounter at startup (which starts recording value changes)
• create other indexes at startup

Shutdown:

• Handle final reporting events
• Handle "partial pertiods" at the end of the simulation for reports that run at regular periods
• A hook to inspect remaining plans, callback queues.

[RobertJacobsonCDC]

Just a brain dump of tangentially related ideas what we might not want but that might be worth talking about:

• stop conditions: formalized criteria for when the simulation should complete
• categories of plans. e.g. maybe a "system plan" needs to run even if shutdown_requested.
• Scheduling plans at a pseudo-time: a way of thinking about "start" and "end" as specific concrete but unknown times.