Container image version dictated by Orchestrator, to enable canarying with centrally managed roll forwards and rollbacks

[vorburger]

The current update.sh-based approach, even if improved with things like #32 and #123 (and maybe #124 and #125) or replaced as suggested in #109, ultimately still leaves the "control" of which version to run at the leaf nodes of the network.

Perhaps a more radical future direction you may want to eventually consider to really scale this up more massively could be to let (something like, or a part of) the Orchestrator decide which Node should run what version?

That would allow it to decide where to attempt "canarying" the next great versions, and gradually roll it out - and centrally roll-back.

How exactly this could be concretely best implemented would require further thoughts.

This may only become more relevant with hundreds or thousands of Nodes than the current tens.