context

we would like to be more resilient to restarts so that we can push concurrency more heavily under load (which often triggers concurrency errors that crash a channel and trigger an auto-restart)
this MR introduces logic for auto-restarts that compartmentalizes each restart to the socket shard in which the channel failure occured
upshot: the system is more resilient to restarts b/c fewer channels overall are affected by any given restart

changes:

if healthchecks fail, group fatal failures by socket id, pass them to refactored _restartAndNotify (accepts socket id and failed channel number), which passes them to refactored restart
restart now does not shut down entire app, merely unsubscribes from channels in shard, aborts correct signald instance, restarts correct socket pool (using new convenience methods on app), then re-subscribes to channels
there is some fancy lodash footwork involved. perhaps we can simplify later! :)
side-effect: make healthchecks happen less frequently in dev to get cleaner logs