Staging/Running Architecture

[minecrawler]

Currently, sim-ecs tries to prepare all resources as early as possible, which means instancing registered types right from the start. This makes things, like adding systems later on, hard because it means keeping track of old and new instances and where they are in use.

That's why I want to change to a 2-phase architecture, where we only work with prototypes in the staging part, and when we start a run, all of the initialization is done and we work with instanced objects, referenced by their types (as we do it at the moment)

I see a few benefits:

• we could still initialize early on, but inside a dedicated cache (I'm not a fan of this, but it would work the same as it is working at the moment)
• we could give each run a separate set of instances to entirely separate even parallel runs
• run-info can easily be transmitted over the wire (or to another thread/worker)
• the staging phase can be modified anytime, and a new run can be initialized after changes
• by comparing run-info, we could even patch a run with new additions (e.g. systems) without a restart (at the moment, we don't track that and re-fill all entity caches with maintain())

The API-impact would be small, which is always a good thing