fix(beholder,anatomist): resolve main-content selectors by priority, not DOM order

[YusukeHirao]

Summary

• Fix main-content resolution in @d-zero/beholder and @d-zero/anatomist: both resolved a priority-ordered selector list via a single querySelector(selectors.join(',')) call, which matches by DOM document order, not by the array's intended priority. An ancestor wrapper matching a lower-priority selector (e.g. #contents) could win over a descendant matching a higher-priority one (e.g. #main) just because it appeared earlier in the DOM.
• Found while manually verifying @d-zero/anatomist against real crawled pages (sampled from nitpicker archives of public corporate sites): a real-world article page had its main-content resolved to div#contents — which wraps both the article body (div#main) and the sidebar (div#aside) — instead of div#main alone, causing the layout classifier to see a 2-column region and misclassify it as complex-grid instead of the article body's actual vertical-stack.
• Both packages now query each selector individually, in array order, so the array's order is the actual priority (matching the existing fallbackSelectors loop pattern in both files).
• Added regression tests to both packages' specs covering: priority winning over later DOM order, a nested higher-priority match beating its lower-priority ancestor wrapper (the exact article+sidebar shape above), an invalid selector being skipped without aborting the rest of the priority list, and documented the existing (unrelated, unchanged) DOM-order-first behavior among multiple elements matching the same selector.
• Documented in both files' module JSDoc that this resolution logic is intentionally duplicated (Puppeteer's page.evaluate can't ship closures) and points at the matching spec cases in the other file, so a future fix to one side surfaces as a spec gap in the other instead of a silent divergence.

Test plan

• yarn test — 142 test files / 1862 tests pass
• yarn build — all 29 packages build
• yarn lint — no new errors (pre-existing warnings only, unrelated to this change)
• Manually re-ran @d-zero/anatomist's CLI against the page that surfaced the bug: mainSelector changed from div#contents (misclassified complex-grid, confidence 0.6) to div#main (correctly classified vertical-stack, confidence 0.8)