Load / soak testing + ingestion SLOs + benchmarks

[AKogut]

Context

Prove the async pipeline holds under burst and sustained load, and prevent regressions.

Scope

• Load test harness simulating bursty CI ingestion
• Soak test for worker/queue stability over hours
• Define SLOs (ingest availability, processing lag) + dashboards
• Benchmark gate in CI to catch throughput regressions

Acceptance criteria

• Documented throughput numbers + SLOs; a perf regression fails CI