[savme]

Today the overview dashboard works out each compute host's reserved CPU and memory inside its own queries, so nothing else can use those numbers. This moves the calculation into recording rules, per host and per cell, and points the dashboard at them. That lets the new capacity alerts read the same numbers the dashboard shows.

Test plan

• Each series, evaluated inline against the deployed environment, matches the dashboard's previous capacity numbers for every cell
• The host set picks up every compute host in deployed environments, including a test environment host that doesn't follow the usual naming
• Finished pods and cordoned hosts are left out of the totals
• The management-plane overlay renders, and the rendered dashboard matches its source

Related to #501