Recursive stop implementation.

[johnwhumphreys]

Meeting Principle 3 & 4 - Recursive stop, if A owns B, and A dies or stops, B must also stop. This guarantees that every running actor has an alive parent it can propagate its errors to, in a path to the root.

Implementation: “manager” actor

This pushes supervision trees down from PythonActor into the core of hyperactor

Every spawn of an ActorMesh also spawns a child “manager” actor on the owner proc which encapsulates the status monitoring, and generates the forwarding of supervision events

[johnwhumphreys]

A diff is out for this and it is working for both procmesh, hostmesh, and actormesh. This works in any order; e.g. if a proc mesh is spawned and actors are spawned after that, they can stop in any order and still work. We spawn a specific actor who's only job is to help shutdown on a mesh. Items know if they have already been stopped. If you shut down a host mesh, it'll stop all procs which stop all actors.

[johnwhumphreys]

@jayasi @keyan @Ritesh1905 - this may be relevant to something else you were working on.

[dulinriley]

Specifically #1941, merging now