Adds the metadata pipeline that makes the gallery block self-curating.
Per-image metadata now resolves through a four-tier waterfall, first
non-empty source wins:
1. Frontmatter (items[].caption etc.)
2. Sidecar file next to the image - tries six patterns:
<name>.yaml / .yml / .json and <basename>.yaml / .yml / .json,
looked up in both site assets and the page bundle (covers album
mode and page-resource mode alike).
3. Embedded IPTC/EXIF via Hugo's .Exif.Tags - caption from
Caption-Abstract / ImageDescription / XPSubject / Description,
credit from By-line / Artist / Creator. Wrapped in `try` so
non-EXIF formats and bad metadata never break the build.
4. Humanised filename - alt fallback only; captions stay empty
rather than fabricated from filenames.
Also adds a third source mode - content.page_resources: true -
auto-discovers image resources attached to the current page bundle.
Drop-in for blog posts and project pages where photos live next to
index.md; sidecar lookup works inside the bundle too.
Fixes a leading-slash quirk: Resource.Name returns "/media/..." for
site assets but a bare path for page-bundle resources; the path is
now normalised so sidecar lookups land in either layout.
Test site adds two sidecar YAML files (04-cabin.jpg.yaml and
06-mountain.jpg.yaml) - the masonry album demo now picks up titles,
markdown captions, and links from those sidecars without listing any
items[].
First-class image gallery block. Two layouts ship today - grid (uniform
tiles, configurable columns 1-6, aspect-ratio presets) and masonry
(CSS-columns Pinterest packing). Click-to-zoom lightbox built in Preact,
portalled to <body> so it escapes the section's `isolation: isolate`
stacking context; keyboard navigation (← → Esc), touch swipe, and body
scroll-lock included.
Two source modes: explicit items[] for curated galleries with per-image
captions/credits/links, or album: <folder> for drop-in folder loading
sorted by filename. Remote URLs (Unsplash/Pexels/CDNs) are passed
through unprocessed; local images get the full responsive variants +
inline base64 LQIP placeholder for instant first-paint over the eventual
sharp image.
Captions, titles, and credits are rendered through Hugo's RenderString
at build time (display: inline) - users get full Markdown (bold/italic/
code/links) in any of those fields with zero JS markdown parser. Alt
text falls back to plainified caption → humanised filename so screen
readers always have something useful.
Caption positions: below (default - accessibility-first), overlay
(magazine gradient), hover (clean grid reveal), none (lightbox-only).
Block-scoped CSS in the co-located style.css is picked up by the
existing block-CSS auto-discovery.
Test site: two demos under home.yaml (curated grid + auto-loaded
masonry) with 9 sample Picsum images stored under
test/assets/media/albums/gallery/ at varied aspect ratios.
George Cushen