split collector and checks needed for every host vs. whole cluster

Some checks need to run once, like checking some value from `ceph -s`. Otherwise, a failing check for say 'nearfull: True` would show up in every single node being checked, which is not necessary.

In order to have split checks, the collector needs to be reworked as well to split *cluster wide* metadata gathering (like `ceph -s`).

- [x] split checks
- [ ] split collector

Start by tracing the collector and the existing split-check implementation described in the issue. Use the cluster-wide `ceph -s` metadata as the example, and consider the work complete when cluster-wide metadata is gathered once while host-specific checks still run for every host.

Written by the indexing model from the issue text.