The preview smoke test failed with "Expected 400 ... got 404" because it
probed /api/info once, ~8s after "Deployment complete". Cloudflare Pages
Functions take a few seconds to propagate; until they are live, /api/*
falls through to the static 404 handler. A live probe minutes later returns
the correct 400.

Replace the single-shot probe with a poll that retries until the function
answers 400, while still asserting the final status is exactly 400 (no
suppression). Applied to the identical preview/staging/prod blocks; bumped
those smoke jobs to a 5m timeout to absorb the worst-case poll.

Co-Authored-By: Claude <noreply@anthropic.com>

Address frontier review on PR #171:

- Add `|| true` to the curl status probe in all three smoke-test jobs
  (preview, staging, prod). Under GitHub Actions' default `bash -e`,
  a connection-level curl failure (DNS not ready / connection refused
  during Pages Function propagation) made the assignment exit non-zero
  and aborted the whole step before the retry loop could run.
- Skip the sleep on the final attempt and emit an accurate 'giving up'
  message instead of a misleading 'retrying...', avoiding a wasted
  10s/15s wait right before the hard failure.
- Render an empty probe result as 'none' for readable logs.
- Bump smoke-test-staging timeout 5 -> 6: it uses the longest curl
  retries (--retry 10 --retry-delay 15) plus the new 15s poll, so its
  worst-case wait exceeds the preview/prod jobs that were bumped 3 -> 5.

Verified locally: loop survives simulated connection failures under
set -e, reaches 400, and the all-fail path exits 1 after exactly 5
sleeps (no wasted final sleep). YAML validated.

Co-Authored-By: Claude <noreply@anthropic.com>