Mock a slow API

Loading states are hard to see on localhost, where every response comes back in a millisecond — which is why so many of them are subtly broken in production. Each route here waits a fixed, known amount before answering, from instant to long enough to trip a default client timeout, so skeletons, spinners, cancellation, and timeout handling can all be checked deliberately.

{
  "latency_ms": 0,
  "message": "Answered immediately."
}