[Epic] Stress & performance benchmark suite

Harper's stress and performance benchmark suite: the workloads we run to catch regressions before customers do, and the gaps in what they currently cover.

## Scope

**In:** the stress/YCSB workloads and their assertions — throughput baselines, p99 ceilings, memory-growth bounds, and the benchmark harness itself.

**Out:** the nightly CI infrastructure that runs them (#1671) and flaky-test triage (#1655). A benchmark that fails because it's flaky belongs there; a benchmark we don't have belongs here.

## Why

Several of these were filed after a real incident exposed a workload we weren't measuring — #1199 and #1197 both cite specific customer situations. That's the pattern worth institutionalizing: a production surprise should leave behind a benchmark, and those benchmarks should live somewhere visible rather than scattered.

_Created during backlog triage. Child issues keep their own priorities and milestones._

The issue names no files or specific child benchmark; start by reviewing the stress/YCSB workloads and benchmark harness, then inspect the child issues and the boundaries with #1671 and #1655. Done is not defined at epic level; it requires scoped child work covering the stated workloads and assertions.

Written by the indexing model from the issue text.