Reuters /technology/ renders a fixed ~20-article list with no lazy-load
or load-more — on high-volume days the window fully churns between two
harvest runs, making gap articles structurally un-capturable from the
DOM. Two recipe-gated additions to fetcher-helper:

- sitemap discovery (sitemapUrl/sitemapFromStep/sitemapPages/sitemap-
  Include): archive-mode link collection reads Google-News sitemap pages
  via same-origin fetch() inside the site session (DataDome cookies
  apply), newest-first, section-prefixed so a site-wide sitemap cannot
  swallow sibling channels' articles. Any failure falls back to the DOM
  scan. Enabled for all four reuters channel recipes (machine-local).
- datelineFallback: wire-service bodies (Reuters) render in wildly
  varying DOM structures, so selector lists go stale and articles look
  paywalled while the full text sits in innerText. When selectors
  under-produce (< bodyMinChars), slice innerText from the dateline
  lead ("AMSTERDAM, Nov 12 (Reuters) -") through the byline
  ("Reporting by"), stripping boilerplate lines — immune to markup
  churn. Also adds cdpEvalAsync (awaitPromise) for in-page fetch().