Split monitoring into short-term / long-term components

The long-term component would require more storage but lower redundancy. This would keep the short-term instances smaller and more performant.