feat(search): Enter bleibt auf Seite + robustere Thumbnail-Erkennung

Startseite:
- Enter/Return löst die Suche jetzt sofort aus (cancelt den Debounce),
  navigiert aber NICHT mehr auf /search. Der Anwender bleibt auf der
  gleichen Seite mit Inline-Ergebnissen.

Thumbnail-Enrichment (searxng.ts):
- Regex-basierte og:image-Extraktion durch linkedom-parseHTML ersetzt.
- Neue Fallback-Kette (in dieser Reihenfolge):
    1. <meta property/name = og:image | og:image:url | og:image:secure_url
                           | twitter:image | twitter:image:src>
    2. <link rel="image_src" href="...">
    3. JSON-LD image (auch tief in @graph; "image" als String, Array,
       Objekt-mit-url)
    4. Erstes <img> in article/main/.entry-content/.post-content/figure
- Relative URLs werden gegen die Seiten-URL zu absoluten aufgelöst
  (z.B. /uploads/foo.jpg → http://host/uploads/foo.jpg).
- maxBytes von 256 KB auf 512 KB angehoben, damit JSON-LD-lastige
  Recipe-Seiten nicht mitten im Script abgeschnitten werden.

Tests (97/97):
- Neu: JSON-LD-Image-Fallback-Test.
- Neu: Content-<img>-Fallback-Test mit relativer URL, die zur
  absoluten aufgelöst wird.

src/lib/server/search/searxng.ts

@@ -1,4 +1,5 @@
 import type Database from 'better-sqlite3';
+import { parseHTML } from 'linkedom';
 import { listDomains, normalizeDomain } from '../domains/repository';
 import { fetchText } from '../http';
 
@@ -77,18 +78,102 @@ function looksLikeRecipePage(url: string): boolean {
   }
 }
 
-const OG_IMAGE_RE =
-  /<meta[^>]+(?:property|name)=["']og:image(?::url)?["'][^>]+content=["']([^"']+)["']/i;
-const OG_IMAGE_RE_REV =
-  /<meta[^>]+content=["']([^"']+)["'][^>]+(?:property|name)=["']og:image(?::url)?["']/i;
-const TWITTER_IMAGE_RE =
-  /<meta[^>]+(?:property|name)=["']twitter:image["'][^>]+content=["']([^"']+)["']/i;
-
-function extractOgImage(html: string): string | null {
-  const m = OG_IMAGE_RE.exec(html) ?? OG_IMAGE_RE_REV.exec(html) ?? TWITTER_IMAGE_RE.exec(html);
-  if (!m) return null;
+function resolveUrl(href: string, baseUrl: string): string | null {
   try {
-    return new URL(m[1]).toString();
+    return new URL(href, baseUrl).toString();
+  } catch {
+    return null;
+  }
+}
+
+function imageFromJsonLd(data: unknown): string | null {
+  if (!data) return null;
+  if (Array.isArray(data)) {
+    for (const d of data) {
+      const img = imageFromJsonLd(d);
+      if (img) return img;
+    }
+    return null;
+  }
+  if (typeof data !== 'object') return null;
+  const node = data as Record<string, unknown>;
+  if (Array.isArray(node['@graph'])) {
+    for (const d of node['@graph']) {
+      const img = imageFromJsonLd(d);
+      if (img) return img;
+    }
+  }
+  const image = node.image;
+  if (typeof image === 'string') return image;
+  if (Array.isArray(image) && image.length > 0) {
+    const first = image[0];
+    if (typeof first === 'string') return first;
+    if (first && typeof first === 'object' && 'url' in first) {
+      const url = (first as Record<string, unknown>).url;
+      if (typeof url === 'string') return url;
+    }
+  }
+  if (image && typeof image === 'object' && 'url' in image) {
+    const url = (image as Record<string, unknown>).url;
+    if (typeof url === 'string') return url;
+  }
+  return null;
+}
+
+const META_IMAGE_KEYS = new Set([
+  'og:image',
+  'og:image:url',
+  'og:image:secure_url',
+  'twitter:image',
+  'twitter:image:src'
+]);
+
+function extractPageImage(html: string, baseUrl: string): string | null {
+  try {
+    const { document } = parseHTML(html);
+    // 1. OpenGraph / Twitter meta tags
+    for (const m of Array.from(document.querySelectorAll('meta'))) {
+      const key = (m.getAttribute('property') ?? m.getAttribute('name') ?? '').toLowerCase();
+      if (!META_IMAGE_KEYS.has(key)) continue;
+      const content = m.getAttribute('content');
+      if (!content) continue;
+      const resolved = resolveUrl(content, baseUrl);
+      if (resolved) return resolved;
+    }
+    // 2. <link rel="image_src">
+    const link = document.querySelector('link[rel="image_src"]');
+    if (link) {
+      const href = link.getAttribute('href');
+      if (href) {
+        const resolved = resolveUrl(href, baseUrl);
+        if (resolved) return resolved;
+      }
+    }
+    // 3. JSON-LD image (Recipe schema etc.)
+    for (const s of Array.from(document.querySelectorAll('script[type="application/ld+json"]'))) {
+      try {
+        const data = JSON.parse(s.textContent ?? '');
+        const img = imageFromJsonLd(data);
+        if (img) {
+          const resolved = resolveUrl(img, baseUrl);
+          if (resolved) return resolved;
+        }
+      } catch {
+        // malformed JSON-LD — skip
+      }
+    }
+    // 4. First content image in article/main
+    const contentImg = document.querySelector(
+      'article img[src], main img[src], .entry-content img[src], .post-content img[src], figure img[src]'
+    );
+    if (contentImg) {
+      const src = contentImg.getAttribute('src') ?? contentImg.getAttribute('data-src');
+      if (src) {
+        const resolved = resolveUrl(src, baseUrl);
+        if (resolved) return resolved;
+      }
+    }
+    return null;
   } catch {
     return null;
   }
@@ -104,8 +189,8 @@ async function enrichThumbnail(url: string): Promise<string | null> {
   if (cached && cached.expires > now) return cached.image;
   let image: string | null = null;
   try {
-    const html = await fetchText(url, { timeoutMs: 4_000, maxBytes: 256 * 1024 });
-    image = extractOgImage(html);
+    const html = await fetchText(url, { timeoutMs: 4_000, maxBytes: 512 * 1024 });
+    image = extractPageImage(html, url);
   } catch {
     image = null;
   }