[warwickschroeder]

What this adds

The audit instance's SQL Server and PostgreSQL retention sweep now emits retention instruments.

• sc.retention.cycle_duration_seconds and sc.retention.consecutive_failures_total on the audit meter, with the same descriptions, buckets and result values as the error instance.
• No rows_deleted_total on the audit instance, because PostgreSQL drops whole partitions and a dropped partition reports no row count.
• One isolated pass per table, tagged retention.entity as audit_messages or saga_snapshots, so a failing table no longer stops the other.
• A pass that leaves rows because its table is too busy records failed, as the Audit retention check already counts it.
• A sweep that fails before its passes records failed for both, the way an unreachable database shows on the error instance.
• Docs in the Retention section of docs/telemetry.md.

Tests

• ServiceControl.Audit.Persistence.Tests/EFCore/AuditRetentionMetricsTests (new): the instrument names, kinds, unit, buckets and meter, and how a cycle records success, failure and shutdown.
• ServiceControl.Audit.Persistence.Tests/EFCore/EFRetentionMetricsTests (new): on SQL Server and PostgreSQL, a sweep records one result per table, keeps the passes isolated, reports deferral, early failure and shutdown, and records nothing when skipped for the lock.
• ServiceControl.Audit.Persistence.Tests/EFCore/EFRetentionCustomCheckTests: a failing pass in either table turns the check red after three sweeps.

[johnsimons]

Looks good, the only thing that may be nice is to also update the dashboard of Grafana in https://github.com/Particular/ServiceControl/blob/master/tools/testing-tool/TestingTool.AppHost/obs/grafana/dashboards/testing-tool.json

[warwickschroeder]

Looks good, the only thing that may be nice is to also update the dashboard of Grafana

Will add to this PR. Thanks