Portrait layout support for AnimeScene/ComparisonCard/KPIGrid + HeroTitle word-wrap

[sethshoultes]

Summary

Four component fixes, developed while producing portrait (1080x1920) short-form video content:

• AnimeScene: new fit="cover" | "contain-blurred" prop. Portrait source images (e.g. a host avatar) rendered with the default cover fit on a landscape canvas get scaled ~2.5x and can push a face off-frame. contain-blurred renders a blurred/darkened cover-fit backdrop plus the full undistorted image letterboxed on top. Also fixes a real bug: two Img elements as direct siblings inside one AbsoluteFill (a flex container) compete for layout space and can both collapse to invisible — each image layer now gets its own nested AbsoluteFill, matching how the existing vignette/particle overlay layers already stack.
• HeroTitle: wraps at word boundaries instead of mid-word (previously split per-character with no word grouping, so long titles could wrap in the middle of a word), scales font size down for long titles instead of overflowing to a third line, and exposes textColor/accentColor/subtitleColor/veil as props instead of hardcoded values.
• ComparisonCard: portrait-mode layout (stacks the two sides top/bottom with a horizontal divider instead of left/right with a vertical one, scales down the divider dimension and caps the value font size), plus cardBackgroundColor/leftColor/rightColor props for dark-theme use.
• KPIGrid: derives grid dimensions from the actual video's width/height instead of a hardcoded 1920x1080, and switches to a single-column layout for portrait.

Explainer.tsx/Root.tsx/SCENE_TYPES.md carry the prop plumbing and documentation for the above.

Test plan

• Rendered multiple 1080x1920 portrait shorts using these components in production (host avatars via contain-blurred, KPI dashboards, comparison cards)
• Verified landscape rendering is unaffected (all new behavior is behind the portrait/contain-blurred branches, existing default paths unchanged)