if printer no longer on network remove from map

[jessfraz]

No description provided.

[paultag]

during the great refactoring I made two changes to deal with this:

1. Discovery will retain a handle to the pool of connected machines which it can modify -- removing and adding machines (access wrapped around a Mutex, so you won't kill an in-progress build)

2. I added a new healthcheck endpoint each Control traited object implements, so we can differentiate between "The API threw an error" and "The printer went away" without solely relying on timeout. For static (non-discoverable) machines (currently just moonraker?) that can fail, it may need a few small tweaks (retry all un-connected but configured machines on a clock, connect, run healthcheck, if good add, if bad, abort).

I'd love to add some logic where, after an Error is dropped from a Machine, we'll re-check health before moving on so that we can drop unhealthy machines that we can't talk to anymore or something

(which is to say, thinking about this one!)