js: implement HTMLMediaElement.currentTime/play/pause/paused

[tannevaled]

Summary

• This engine has no <video>/<audio> support at all — no entry anywhere in isReplacedTag or any binder file.
• Implemented the storage-only subset a script needs to stay internally consistent: currentTime (read/write), play()/pause() (methods toggling paused), the same "storage without real playback effect" shape as scrollTop/scrollLeft — no real media decoding/playback, since a static renderer has no timeline to advance.
• Found by sweeping a third, freshly-fetched corpus page (archive.org's own item-detail viewer) after both prior corpora (this session's ten-page bench set, then github.com's own bundles) were exhausted.
• Most of the initial grep hits (duration, volume, muted, .paused) turned out to be jQuery's own $.fx.speeds animation internals and Bootstrap's carousel widget — unrelated to media elements at all. Traced each hit back to its actual surrounding code before trusting it. The one confirmed real trigger: clicking a transcript entry finds the real <video> element (shadowRoot.querySelector("video")) and does videoEl.currentTime = ...; videoEl.play() to seek and resume playback.
• play() returns a real Promise, matching the spec's own return type. duration/volume/muted/readyState/playbackRate deliberately left out — no confirmed real trigger for any of them survived closer inspection.

Test plan

• New TestMediaElementCurrentTimeAndPlayPause, git-stash-confirmed: reverting turns currentTime/paused into undefined and makes .play() throw TypeError: Object has no member 'play'.
• go build ./... && go vet ./... && go test ./... clean.
• Coverage floors held (css 99.5%, layout 100%, paint 100%, dom 98.4%, paginate 100%).
• Bench vs. real headless Chrome: exactly flat on pkg.go.dev (0.716, unchanged) as expected, no visual effect by design. caniuse.com showed unusually slow fetch times across three runs — investigated rather than assumed: an isolated re-run showed real Chrome taking even longer on the same fetch, confirming a real but purely environmental/network condition, not a code regression; SSIM stayed within its own established noise band throughout.

🤖 Generated with Claude Code