Publish a repeatable alpha performance baseline for CRUD, ingest, MQTT, and RPC

[jillesvangurp]

Summary

PRs #32 and #33 improved replay throughput and reduced downstream delivery overhead, but issue #2 still lacks a repository-level performance baseline that can be rerun intentionally.

Scope

• define representative repeatable scenarios for CRUD, ingest, MQTT publication, and RPC
• capture concrete throughput and latency measurements for the current hub architecture
• document how engineers rerun the performance checks locally or in CI-adjacent workflows
• publish current limits and regression thresholds or comparison guidance

Why

This is the remaining measurable-baseline work from issue #2 after the April 2 performance improvements landed.

Part of #2
Related to #32
Related to #33

[jillesvangurp]

PR #44 lands the benchmark tooling, replay dataset/docs, drop-counter OTEL metrics, and the committed fence hot-path reduction from this slice. It advances the repeatable performance-baseline work, but it does not yet establish the full CRUD/ingest/MQTT/RPC baseline described in this issue, so I left this issue open.