Clouds: define and support liveness/readyness checks

Kubernetes (and probably other clouds as well) support a liveness and readyness check. These currently map on the Docker healthcheck (status = OK?), but we might want to support a separate liveness check. For instance, when you do a rolling upgrade but use BBolt, you might want to stop the old instance when the node is "live" (no BBolt connection yet) rather than "ready" (when the BBolt connection is open), because BBolt doesn't support multiple connections to the same DB.

Start by tracing how Kubernetes liveness and readiness checks currently map to the Docker healthcheck and how rolling upgrades use those states. Examine the BBolt connection constraint described in the issue, then define the separate check behavior and verify that an old instance can stop at liveness without waiting for readiness.