Speed up main page load and show explicit loading/error states

[caufieldjh]

Fixes #660.

The main page fetched the full ~30 MB registry/kgs.jsonld on every visit and parsed it before rendering anything, while the count badges were hardcoded to 0 resources / 0 KGs. For several seconds the registry looked empty, with no indication that data was still loading. A failed load reset the badges to 0 too, so an error looked almost identical to both "still loading" and "registry is empty".

This addresses all three items in #660.

1. Improve data loading speed

The table only reads ~13 fields per resource (id, name, description, category, domains, collection, taxon, activity_status, evaluation_page, homepage_url, license, publications) — the rest of the 30 MB is contact/product/publication detail the main page never renders.

New util/make_summary_json.py generates registry/kgs-summary.json with just those fields, wired into the Makefile next to the existing kgs.jsonld target. The main page fetches that instead:

	Raw	Gzipped (what Pages serves)	Fetch + parse (localhost)
kgs.jsonld (before)	28.5 MB	1.75 MB	~880 ms
kgs-summary.json (after)	0.67 MB	0.16 MB	~30 ms

~11x less to download, and the slow synchronous JSON.parse of a 30 MB string is gone. The full kgs.jsonld is untouched and still offered for download / API use; custom.js falls back to it automatically (with a console warning) if the summary is missing, e.g. on a deployment built before this change.

2. Loading indicator

The badges and table area now ship a Loading registry data… spinner in the static markup, so it's visible from first paint rather than only after the fetch resolves (the old Loading resources... only appeared after the fetch already succeeded, so it never covered the actual slow part). Counts appear only once data has loaded, and No resources found now shows only when a real filter returns nothing.

3. Make errors obvious vs. still-loading

A failed load now shows a distinct error alert that names the cause, states explicitly that this is a load failure rather than an empty registry, and offers a Retry button. Badges go to — with bg-danger instead of 0, so failure can't be confused with loading or emptiness.

A failed taxon_mapping.yaml fetch (previously console-only, invisible) now shows a dismissible warning that taxon filtering is degraded to exact matches; the table itself is unaffected.

Also fixed (found while investigating — the "subtle error" from #660)

• The <noscript> fallback was broken: _includes/resource_table.html iterated {% raw %}{% for ont in site.ontologies %}{% endraw %}, but the data key is site.resources, so no-JS users got an empty table plus OBO-style links (purl.obolibrary.org, Ontobee) irrelevant to this registry. Replaced with a short notice + data download links. Deliberately kept small: <noscript> content is downloaded by every browser even with JS enabled, so rendering 1000+ rows there would add page weight for everyone and undercut the point of this PR.
• jQuery was loaded twice with different versions — 3.6.1 in index.html, then 3.6.0 again in _includes/scripts.html. custom.js is the only jQuery consumer in the repo and index.html loads its own, so the duplicate is dropped.

Verification

I drove the real custom.js against the real registry data over HTTP in jsdom (real jQuery 3.6.1, real DOM, real sorttable.js). 31/31 checks pass, covering:

• loading spinner visible while data is in flight; badges say "Loading", never 0
• successful render (1012 rows), correct counts (1012 resources / 153 KGs)
• resource-type, taxon (hierarchy-expanded), and search filters all still work on the slim data
• empty search result shows the info "no resources found", not an error
• error state: distinct alert, — badges with bg-danger, Retry button present
• Retry recovers and renders the full table with normal badges
• summary-missing → transparent fallback to kgs.jsonld
• taxon-mapping failure → visible warning, table still renders

I also diffed the summary against kgs.yml field-by-field across all 1012 resources: identical values, and zero rendering differences for the publication button (the 111 resources whose publications count differs all have empty lists, which render the same disabled button either way).

Not covered: no local Jekyll install, so the full site build isn't exercised here — but resources is confirmed as a top-level key in _config.yml, so the {% raw %}{{ site.resources | size }}{% endraw %} in the new noscript block resolves.

🤖 Generated with Claude Code