docs: outbox sweeper health metrics

[iancooper]

Documents the outbox sweeper health metrics added in BrighterCommand/Brighter#4562 (issue BrighterCommand/Brighter#4560, ADR 0081-timed-outbox-sweeper-health-metrics).

contents/Telemetry.md: new "Outbox Sweeper Metrics" section

• Turning the metrics on with AddBrighterInstrumentation().
• The three instruments (tick.lag, sweep.duration, sweeps) and the meaning of each outcome.
• How the sweeper keeps time on its own thread, and the limit that a sweep's async I/O still runs on the pool.
• Suggested PromQL alerts: sweeper stalled, sweeps starting late, sweeps failing. The stalled alert sums across instances, because with a distributed lock only one instance completes sweeps.
• Histogram buckets: the SDK's defaults were designed for millisecond values, and these histograms record seconds.
• A table for telling a slow broker from a starved process, and turning on the System.Runtime meter for thread-pool queue length and thread count.

contents/BrighterOutboxSupport.md

• The TimerInterval row notes that it must be at least 1.
• The sweeper section links to the new metrics section.

Checked:

• The C# snippets were compiled and run against the Brighter branch, with a 1 s interval and an in-memory exporter. All three instruments were exported with the documented types, units and outcome tag. The bucket views applied, and System.Runtime produced the thread-pool metrics.
• linkcheck, pagelint (including --changed origin/master and --plant) and symbolcheck pass.
• The PromQL was not run against Prometheus. The metric names assume the Prometheus exporter's usual renaming.

Merge after BrighterCommand/Brighter#4562.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SpjnqGWbARr3Z3ofGc6C2x