[1/5] perf: ship feature posters as webp instead of jpeg

[ralph-cmyk]

Merge order 1 of 5 — base of the stack; #1106, #1107, #1108, and #1110 all build on this branch.

Problem. The homepage ships 864 KB of JPEG poster images. Twelve load
immediately, because every feature video shows a poster while the video itself
stays deferred.

Why it matters. That is roughly 59% of the homepage's total weight, on the
one marketing page that has to make the first impression. The docs pages score
100 on mobile performance; the homepage scores 72.5.

What changed. reencode-all.sh now exports WebP posters instead of JPEG;
posters regenerated; the poster-path convention in Feature.astro and four
hardcoded references updated. Videos are untouched and still preload="none".

KPI — starting point. Poster payload 864 KB (13 files, du -ch).
Homepage total 1.46 MB, mobile lab Performance 72.5, mobile lab LCP 5.84 s
(PageSpeed medians of 2 runs, 23 Jul 2026). Field LCP p75 1.44 s — already good,
and must stay that way.

KPI — expectation. Converted-poster payload down ~51% to ~420 KB (measured
on 5 sample conversions at q78; the directory keeps one pre-existing 105 KB
webp). Homepage total ~1.02 MB. Mobile lab LCP and Performance should improve;
field LCP p75 must not regress.

How to measure it. du -ch dist/features/posters/* for payload. Re-run
PageSpeed Insights on https://www.rocketsim.app/ (mobile, median of 2 runs)
one week after deploy and compare against 72.5 / 5.84 s / 1.46 MB.

Risk / rollback. WebP is supported by every browser RocketSim targets, and
one poster on the site is already WebP. If a poster renders wrong, revert the
commit — posters are generated artifacts, nothing else depends on them.

[AvdLee]

Replaced the black asc-drag-drop-demo.webp poster: the video fades in from black, so the script's first-frame export was a black image (the old .jpg had the same issue, it was just never noticed). The new poster is hand-picked from 12s (the App Store Connect upload view), same cwebp -q 78 pipeline, native 1440x810, 32 KB. Heads up: re-running reencode-all.sh will overwrite it with the black frame again — worth teaching the script a per-video frame offset or a skip-existing flag at some point.