Drop LWN subscriber-only articles instead of stamping paywalled

LWN's subscriber-only articles have no public version reachable by
archive snapshots or bypass UAs, so stamping them as paywalled produces
clicks that always dead-end. Detect the marker text and skip ingestion
entirely.

internal/ingestion/article.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"net/url"
 	"strings"
 	"time"
 
@@ -68,7 +69,12 @@ type ArticleMeta struct {
 	Fetched    bool   // true if we got an HTTP 200 with HTML
 	FetchError bool   // true if the fetch failed at the network/HTTP layer
 	Paywalled  bool   // true if the page explicitly declares gated access
-	Status     int    // last HTTP status seen (0 on transport error)
+	// SubscriberOnly is true when the source publishes the article in a
+	// form that genuinely has no public version — no archive snapshot, no
+	// bypass UA, nothing. Callers should drop these from view entirely
+	// rather than stamp them as paywalled.
+	SubscriberOnly bool
+	Status         int // last HTTP status seen (0 on transport error)
 }
 
 // Gated reports whether the response carries a strong gating signal: an
@@ -107,6 +113,9 @@ func shouldRetryAsBot(m ArticleMeta) bool {
 	if m.FetchError {
 		return false // transport failure won't be fixed by a different UA
 	}
+	if m.SubscriberOnly {
+		return false // genuinely subscriber-only; no UA changes that
+	}
 	if m.Gated() {
 		return true
 	}
@@ -185,9 +194,27 @@ func fetchArticleMetaOnce(articleURL, ua, referer string) ArticleMeta {
 	meta.ImageURL = extractOGImage(doc, articleURL)
 	meta.BodyChars = extractBodyChars(doc)
 	meta.Paywalled = detectPaywall(doc)
+	meta.SubscriberOnly = detectSubscriberOnly(articleURL, doc)
 	return meta
 }
 
+// detectSubscriberOnly returns true for sources known to publish articles
+// with no public version reachable by any workaround we have (archive
+// snapshots, bypass UAs, etc). Currently only LWN's subscriber-only
+// articles, which display a fixed marker in the article body.
+func detectSubscriberOnly(articleURL string, doc *goquery.Document) bool {
+	u, err := url.Parse(articleURL)
+	if err != nil {
+		return false
+	}
+	host := strings.ToLower(u.Hostname())
+	if host != "lwn.net" && !strings.HasSuffix(host, ".lwn.net") {
+		return false
+	}
+	body := strings.ToLower(doc.Find("body").Text())
+	return strings.Contains(body, "available only to lwn subscribers")
+}
+
 // isCloudflareChallenge reports whether a non-200 response is Cloudflare's
 // bot-block or JS-challenge page rather than a real publisher gate. The
 // signatures are stable: cf-* response headers, the cf-error-details body