Teleport will stay unhealthy upon dropping to zero resources while unhealthy

[espadolini]

Expected behavior:
A temporarily unhealthy App/Kube/DB Teleport agent that goes down to zero matched resources to serve will report as healthy (as nothing can be wrong anymore).

Current behavior:
Since Teleport recovers from being unhealthy if a successful heartbeat happens, when there's no running heartbeats, Teleport will stay unhealthy forever.

This is especially problematic in the teleport-kube-agent chart on a rollout, since the new pod is the one most likely to become unhealthy, resulting in a stuck rollout.

Bug details:

• Recreation steps: cause a Teleport agent to become unhealthy and then delete all dynamic resources matched by the agent; this was initially discovered by changing filters in a discovery service in kubernetes, causing mass deletion of apps (which will also likely cause the agent to report as unhealthy for a bit, because of There is no way to gracefully stop a single inventory control stream heartbeat #50237) and leaving the new agent with no apps to serve.