reader-factory: benchmark single-actor vs factory

[disconsented]

Benchmark the read path single-actor vs factory to prove the win and set the default pool size.

Scope

• Drive concurrent item-detail requests; measure p50/p99 latency and throughput.
• Compare single-actor (current) vs factory at pool sizes 2 / 4 / 8.
• Use the result to pick the default read_workers value.

Why

Without numbers there's no way to know the factory beat the much simpler direct-DB_POOL approach, or what pool size is worth running.

Depends on

• reader-factory: convert ItemActor to a ractor factory #59 — ItemActor → factory

Parent: #51