Add self kill when controller no longer messages

[dulinriley]

Summary:
Fixes: #2198

One of the principles of ownership we want to maintain in monarch is that if the owner of
a resource (meshes of hosts, procs, actors) is not alive, that all of the resources it owned
are also not alive. This should include if the owner hits a hard failure that prevents it from
running any cleanup.

Before this change, we didn't have that guarantee, an actor would keep running if its owner
went away, as long as the actor didn't message the owner.
This change tracks every time a GetState message is received by a ProcMeshAgent, and tracks
which controller sent it. We use this information to infer the controller is alive and well.
We use a new KeepaliveGetState message which is the same as a GetState but has an explicit
side effect. Only ProcMeshAgent uses this for actors at the moment, but we will expand to cover
procs and hosts too.

Benefits of this change:

• Actors might be using resources like CPU, memory, or GPUs that are no
  longer needed and can be recovered.
• If you disconnect the client, all of the actors it spawned
  will shut themselves down. And then any resources those actors spawned will also shut
  themselves down recursively. You don't need to run any special cleanup function on the client.

This feature is hidden behind a flag, and this flag defaults to off. It may have unexpected issues
when we start having actors stop themselves, so we want to test this thoroughly.

Differential Revision: D92726630

[meta-codesync]

@dulinriley has exported this pull request. If you are a Meta employee, you can view the originating Diff in D92726630.