Ref T12298.

• If all daemons in all pools are hibernating already, we may exit immediately without sending EXIT events and fail to update the web UI.
• A max() should be min() ("sleep for at most 180 seconds", not "sleep for at least 180 seconds").
• If there are no daemons in a pool (unlikely/theoretical), we might not set the shutdown flag on the pool correctly.
• An autoscale message may actually mean that a pool is exiting.

• Ran bin/phd debug task, waited for daemon to hibernate, killed it, saw it vanish from daemon console.
• Saw it hibernate for 60s, not 180s.
• Uh, killed a pool real fast and nothing broke? This one is hard/impossible to test.
• Read the newer autoscale message.