Improve LVAlert Overseer configuration
A few ideas:
- Move overseer into its own container so we don't have one instance in each gracedb container (will have to update the overseer code to allow connections from outside localhost (optionally; not by default) and secure it on AWS)
- Add some kind of monitoring to it (in AWS? via some Django URL that nagios can query?)
- Fix the client persistance with the failover to the basic LVAlert client