In order to create a meaningful SLO for the devices in the CrOS
test lab, we need to track some specific metrics relevant to the
managed inventory:
  * Total availability - The number of DUTs in each of the following
    states: working, broken, idle.
  * Working spares - For a given model, the number of working DUTs
    minus the total size of all critical pools for the model.
  * Spares buffer - For a given model, the size of the 'suites' pool.
  * Time spent broken - The time (as a distribution) that all "broken"
    DUTs have spent in that state.
  * Time spent idle - The time (as a distribution) that all "idle" DUTs
    have spent in that state.

Currently, the regular inventory e-mail generates the first three metrics
twice a day, as part of the "model inventory" runs.  The "time spent broken"
and "time spent idle" metrics can be easily calculated.

Ideally, there might be a "time spent working" distribution as well.
However, within the current database and API, calculating "time spent
working" is hard, and there's no identified requirement for the information.