bench: a reproducible benchmark harness with published numbers

The README claims "the performance of Rust". There is no benchmark anywhere in the tree — no
benches/, no criterion target, no harness under examples/.

An unbenchmarked performance claim is worth less than no claim, because the first reader who
cares will run their own and publish it.

Build a harness that lives in the repo and that a stranger can run:

• a single scenario definition (N jobs, fixed payload, fixed concurrency, measured to
  completion, not to enqueue) run against FlexiQ and against Celery, Dramatiq, RQ and BullMQ
• one command, pinned dependency versions, machine and backend recorded in the output
• results as a committed artifact, then a chart on the landing page that links back to the
  script that produced it

Report enqueue throughput, end-to-end latency percentiles, and idle cost separately. A queue
that wins on throughput and loses on p99 should show both.

Publish the losses. A benchmark that FlexiQ wins on every axis reads as a benchmark FlexiQ
wrote.

Start with README and inspect the repository for the currently absent benches/, examples/, and Criterion entry points. Define how one command will run the named queue scenarios and record machine, backend, dependencies, throughput, latency percentiles, and idle cost. Done means a committed result artifact and landing-page chart link back to the producing script, including losses.