fix(css): emit the base reset for tags evicted from the defaultStyle cache (#468)

[dylan1951]

Closes #468.

Problem

generateDedupedBaseCSS built the per-tag base reset by reading the cache directly:

const styles = cache.defaultStyle.get(tagName)
if (!styles) continue

cache.defaultStyle is an EvictingMap capped at MAX_DEFAULT_STYLE = 30, evicting FIFO. The reset is generated at the very end of a capture, so any document using more than 30 distinct tags has already evicted its earliest-registered tags. The lookup returns undefined and continue silently emits no reset rule for exactly those tags.

The base reset is what neutralizes the UA stylesheet inside the foreignObject. Properties the page reset to the CSS initial value get diffed out of the element's generated class precisely because the base reset is expected to cover them — so when it's missing there is nothing left to hold the UA default off: h1..h3/p margins, hr borders, list padding all come back, and the capture reflows taller than the source.

Which tags are hit depends on registration order, so the corruption is silent and varies per page. Observed on a real site: 39 distinct tags, and the tags at first-appearance ranks 15–23 (picture, main, h1, h2, h3, strong, p, hr) all lost their reset while h4 at rank 25 survived. Four <h3>s regained the UA 1em margin, growing the article by 121px and pushing a button 171px down over the section below it.

Fix

Resolve through getDefaultStyleForTag(tagName) instead of reading the map raw. It's memoized, idempotent, and re-derives an evicted entry on demand. NO_DEFAULTS_TAGS still yields {} and is dropped by the existing empty-key guard, so SVG/head tags are unaffected.

One line, plus a comment recording why the raw map read is wrong here.

Cost

Each evicted tag now re-derives once at end of capture — a sandbox element plus a computed-style read. It's a single bounded pass (each tag is visited once, so ~9 derivations on the 39-tag page above), negligible against a full capture.

Raising MAX_DEFAULT_STYLE would also make this particular page work, but it only moves the cap rather than removing the failure mode — the next document with more tags hits it again. Happy to go that route instead if you'd prefer.

Tests

__tests__/utils.css.base-reset-eviction.test.js, 2 cases:

1. end to end — a capture using >30 distinct tags still emits a base reset for an early-registered h3
2. unit invariant — every used tag that has defaults appears in the output, even starting from a cleared cache

Both fail on main.

Note for reviewers: the fillers in the end-to-end test must have distinct UA styling. Tags that compute identically (aside/section/nav/header/…) collide on the style-key memo in styles.js, never call getDefaultStyleForTag, and so never occupy a cache slot — with those the capture registers only 28 tags, stays under the cap, and the bug doesn't trigger.

Full suite: 691 passed / 1 skipped, against 689 / 1 on main — no regressions. npm run lint and npm run test:types clean. Verified on Chromium; I don't have the Firefox/WebKit Playwright binaries locally.