Strip inline scripts from reader body extraction

Datawrapper embed resizers (and other inline scripts) were leaking into
reader mode as literal text. Two extraction paths were affected:

- parser.go extractContentText/extractLede stripped only script tags via
  htmlTagRe, leaving the JS body behind. This is the path that usually
  wins for feeds shipping full content:encoded (e.g. Politico).
- article.go goquery paths call .Text(), which concatenates script source.

Both now drop whole <script>/<style>/<noscript> elements before pulling
text. Paywall detection (JSON-LD) still runs before the goquery strip.

internal/ingestion/article.go

@@ -193,12 +193,18 @@ func fetchArticleMetaOnce(articleURL, ua, referer string) ArticleMeta {
 
 	meta.Fetched = true
 	meta.ImageURL = extractOGImage(doc, articleURL)
+	// Paywall detection reads <script type="application/ld+json"> and the LWN
+	// marker, so run it before we strip non-content nodes below.
+	meta.Paywalled = detectPaywall(doc)
+	meta.SubscriberOnly = detectSubscriberOnly(articleURL, doc)
+	// Strip <script>/<style>/etc. before pulling body text — goquery's .Text()
+	// includes the source of inline scripts (e.g. Datawrapper embed resizers),
+	// which would otherwise leak verbatim into reader mode.
+	stripNonContent(doc)
 	// One body extraction serves both purposes: the text feeds reader mode and
 	// its length is the paywall/thin-body signal.
 	meta.BodyText = extractBodyText(doc)
 	meta.BodyChars = len(meta.BodyText)
-	meta.Paywalled = detectPaywall(doc)
-	meta.SubscriberOnly = detectSubscriberOnly(articleURL, doc)
 	return meta
 }
 
@@ -395,9 +401,19 @@ func FetchArticleBody(articleURL string) string {
 	if err != nil {
 		return ""
 	}
+	stripNonContent(doc)
 	return extractBodyText(doc)
 }
 
+// stripNonContent removes nodes whose text content is never article prose but
+// which goquery's .Text() would otherwise concatenate into the extracted body:
+// inline scripts (Datawrapper/embed resizers, analytics), CSS, and the fallback
+// markup inside <noscript>/<template>. Must run after any logic that inspects
+// these nodes (e.g. JSON-LD paywall detection).
+func stripNonContent(doc *goquery.Document) {
+	doc.Find("script, style, noscript, template").Remove()
+}
+
 func extractBodyText(doc *goquery.Document) string {
 	out := joinParagraphText(doc.Find("article p, main p"))
 	if len(out) < PaywallBodyThreshold {