Componentize, design tokens, earliest-snapshot links

Three independent improvements:

1. Earliest-snapshot wayback links. site_data now builds each
   headline's wayback_url from curated.first_seen_ts instead of
   reusing the enriched.csv link (which used last_seen_ts and
   tended to point at 2024 mementos for old articles). No
   re-fetching needed — the metadata didn't change, only the
   link does. Verified: the 2015 Cubs-Pirates wild-card liveblog
   now links to a 2015-10-09 capture (three days after the game)
   instead of a 2024 capture.

2. Componentized SvelteKit. The home route was carrying all of
   the header, footer, tagline, search, year list, and byline
   teaser inline. Split into seven components under
   src/lib/components/ with a barrel re-export:
     SiteHeader / SiteFooter / Tagline / SearchBox /
     YearList / BylineTeaser / EntryList
   Pages now read top-down at a glance.

3. Expanded :root design tokens in app.css. Colors, font stack,
   font sizes, weights, line-heights, spacing scale, page
   max-width, and rule weights all expressed as custom
   properties. Restyling the whole site is editing a dozen
   tokens.

Also: svelte.config.js now picks base path automatically. CNAME
present in static/ -> custom domain (base ''). Absent -> gh-pages
subpath ('/fakethirtyeight.com'). Drop a CNAME file in to flip.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: palewire

src/fakethirtyeight/site_data.py

@@ -148,21 +148,26 @@ def _build_record(
     kind = curated_row.get("kind") or ""
     url = curated_row.get("url") or ""
 
+    # Always link to the EARLIEST snapshot we know of (curated.first_seen_ts).
+    # The article's metadata is identical across captures, but the earliest
+    # snapshot is closer to "how the site actually looked then" and avoids
+    # linking 2015 articles to 2024 mementos captured during the site's
+    # final months. Falls back to last_seen_ts when first is missing.
+    earliest_ts = (
+        curated_row.get("first_seen_ts") or curated_row.get("last_seen_ts") or ""
+    )
+    wayback_url = _build_wayback_url(earliest_ts, url) if earliest_ts else ""
+
     if enrich_row:
         # Re-clean titles to apply any extractor improvements that landed
         # after enrich.csv was written (e.g., Blogspot-era prefix stripping).
         title = _clean_title(enrich_row.get("title") or "")
         byline = enrich_row.get("byline") or ""
         date = enrich_row.get("published_at") or ""
-        wayback_url = enrich_row.get("wayback_url") or ""
     else:
         title = ""
         byline = ""
-        date = curated_row.get("last_seen_ts") or curated_row.get("first_seen_ts") or ""
-        wayback_url = _build_wayback_url(
-            curated_row.get("last_seen_ts") or curated_row.get("first_seen_ts") or "",
-            url,
-        )
+        date = earliest_ts
 
     # Fall back to a slug-derived title when we couldn't extract one.
     if not title: