Review-cleanup sweep: vendor d3, PaCMAP relabel, footer, data tables, markers, discussions dark mode

[realmarcin]

Stacks on feat/dark-mode-toggle (do not merge before that branch). Additive dataviz/artifact-design review cleanup — light and dark themes both preserved. Templates edited + pages regenerated via just gen-pages (477/477 traits).

Changes

• R1 — Vendor d3 (drop CDN): added pages/d3.v7.min.js (v7.9.0, byte-identical to the shared CommunityMech copy). Pointed umap.html, graph.html, and trait.html script src at {{ root }}d3.v7.min.js. No d3js.org refs remain (355 trait/scatter pages now load the local file).
• R2 — PaCMAP relabel: index.html embedding card body "Interactive UMAP" → "Interactive PaCMAP" (umap.html header already said PaCMAP; data keys umap_x/umap_y and file names left unchanged).
• R4 — Footer: fixed base.html's stale footer (4 → 5 Mechs incl. ProteinTraitsMech; METPO w3id → BioPortal) to match index.html; rendered the current Mech (TraitMech) as plain <span> text (no self-link) in both base.html and index.html.
• R5 — Data-table fallback: keyboard-reachable <details> "View data as table" + Download-CSV link directly below the plot on umap.html and graph.html, built from the same points the scatter renders (label, CURIE, category, x, y), theme-tokenized.
• R6 — Marker size: scatter radius r=4 → r=5 in umap.html and graph.html.
• Discussions dark mode: app/discussions/index.html gets a vendored theme-toggle.js, a dark palette (OS default + data-theme override), and a reduced-motion block.

Preserves the HREF_BY_ID click-through fix already on the base branch.

Verification

• grep d3js.org over src/ + pages/: 0 code refs remain.
• User-facing "UMAP" left where projection is PaCMAP: 0.
• Footer TraitMech self-link: 0 (plain <span> everywhere).
• <details>/table + r=5 present on both scatter pages.
• Light + dark token blocks untouched (additive edits only).

🤖 Generated with Claude Code

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

[realmarcin]

Reviewed the PR's own diff (base feat/dark-mode-toggle): base.html, index.html, trait.html, umap.html, graph.html, style.css, discussions.

Clean — no blocking issues. All claims verified:

• Vendored d3 at pages/d3.v7.min.js (v7.9.0); {{ root }}d3.v7.min.js resolves correctly at every depth (rendered trait pages → ../../d3.v7.min.js, umap/graph → d3.v7.min.js). No d3js.org <script src> refs remain in rendered pages (only the vendored file's own header comment matches).
• Footer now 5 Mechs incl. ProteinTraitsMech, METPO → BioPortal, TraitMech as plain <span> (no self-link) — base.html matches index.html. The proteintraitsmech lowercase URL is correct (repo is lowercase; lowercase Pages URL returns 200, CamelCase 404).
• Data table: builds from the same points; label/CURIE/category/x/y with HTML-escaped cells, linked labels via HREF_BY_ID, and a data-URI CSV with proper RFC-4180 quoting. r=5 markers on both plots. PaCMAP relabel on the index card.
• Render consistent: 477/477 trait pages, r=5 and vendored d3 present in rendered umap/graph.

Minor, non-blocking (out of stated scope): trait.html's causal-graph section still loads dagre from cdn.jsdelivr.net, so those pages retain one CDN dependency even though d3 is now local. Fine for this PR (scope was d3), but worth a follow-up if the goal is fully offline pages.