Write a capacity plan for doubling the fleet

The notebook is three questions in order: how much of each bucket does each option use, how much waiting does that utilisation cost given how bursty the traffic is, and how many workers does the remaining synchronous load need. Inputs live in inputs.json, so changing an assumption is a one-line edit and a re-run. The decision that matters most is the utilisation target: it comes from burstiness and from what else needs headroom (failover, forecast error), not from the average.