I had a request from Chad if it was reasonable to collect better usage statistics from gracedb and lvalert. This seems totally doable, but would just require some parsing/archiving from the gunicorn access logs. The archiving schema is changing with the new deployment, but it shouldn't be that hard to implement after docker-swarm comes up and running.

Note that this ticket has some overlap with #176, so maybe this can be a "two-birds" thing once I get it running.