More HostManager "restart policy" control

[mhasself]

Agent instances should have more "manage" options. Right now it's either "yes" (always keep the agent up) or "docker" (basically the same thing but in a docker container), or "no" (ignore). We need at least a mode where the agent can be started / stopped through HostManager, but is not aggressively managed.

Use cases:

• On smurf server there are lots of agents, but other things (e.g. jackhammer) need to control the life cycle sometimes, without getting interfered with.
• Agents that are in development, or hardware that is only intermittently available; such instances should not always be forced up on HostManager restart.