Operational metrics for portunus, backed by seshat (one counter set per node, per cluster).

Gauges are published per node by the machine's handle_aux/5 on each tick (set_gauges/2), so every member reports its own view, not the leader's. is_leader and leader_changes_total come from the machine's state_enter/2, which fires on every member. Counters are bumped via incr/2 from the leader's apply effects, never inside apply/3, which runs on every replica and would N-fold over-count. The one exception is failures_due_to_lack_of_online_quorum_total, bumped on the client side when a command or query cannot be served because the Raft cluster has no online majority, and so never reaches the machine.