Add error-side alert entries for the periodic kinds so a failing schedule is distinguishable from a stopped one

The five processed-side liveness entries issue #1277 (closed) adds to docs/dev/alerting.md read jobs_processed_total{kind=<kind>, status="ok"} and nothing else. A tick that runs and fails books status="error" and never ok, so each of those rules fires for a failing schedule exactly as it does for a stopped one, and no rule on the page reads the error side of any of the five kinds.

The error series exists today. InitMetrics seeds ok, error and discard at zero for every registered kind when the River client is built (internal/jobsriver/register.go:211-229), so these entries need no new instrumentation and can land independently of the enqueue-side work in #1704 (closed).

This issue is split from #1277 (closed).

Scope

  • Error-side alert entries in docs/dev/alerting.md for the five periodic kinds #1277 (closed)'s liveness entries cover: the three drain triggers, the reconciliation orphan sweep trigger and the remote health sweep (internal/accounting/register.go:39-52, internal/lifecycle/register.go:36, internal/remote/healthjob.go:400).
  • Each entry reads jobs_processed_total{kind=…, status="error"|"discard"}, in the shape of the page's existing lifecycle:tombstone-sweep error rule (docs/dev/alerting.md:96, s3, alert_type: cause, for: 30m), or one entry selects the five kinds in a single kind=~ selector.
  • The severity, window and for: chosen to carry the readings the page already documents: a failed attempt books status="error" and will retry (the drain triggers take River's default 25 attempts, the two reconciliation kinds 3, the health sweep 1); a spent budget books status="discard"; and a deploy landing inside a tick force-cancels the kinds without CancelOnDrain and books error (docs/dev/alerting.md:184).

Acceptance

  1. Each of the five kinds has an error-side entry on the page, or one entry selects all five.
  2. A failed attempt and a spent budget are each readable from the page, so an operator paged by a stopped schedule's liveness rule can tell a stopped schedule from a failing tick.
  3. No Go file, metric or spec changes: the series already exists and this work is documentation.

This is a bot message 🤖 — /smurfit