fix: serialize article fetches in lookup to avoid lxml double-free

trafilatura.extract dispatched via run_in_executor isn't safe to run
concurrently — two parallel calls can crash the process with a
libxml2-level double free. The top-level Wikipedia+SearXNG gather is
fine; only the inner per-article extraction needs to stay sequential,
matching the pre-parallelization behavior.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/fabledassistant/services/tools/web.py

@@ -68,20 +68,23 @@ async def lookup_tool(*, user_id, arguments, **_ctx):
 
     web_payload: list[dict] = []
     if search_results:
+        # Sequential fetches: trafilatura/lxml is not safe to run concurrently
+        # via run_in_executor — parallel calls can trip a libxml2 double-free.
         from fabledassistant.services.rss import _fetch_full_article
 
-        top = [r for r in search_results[:2] if r.get("url")]
+        for r in search_results[:2]:
-        contents = await asyncio.gather(
+            url = r.get("url", "")
-            *(_fetch_full_article(r["url"]) for r in top),
+            if not url:
-            return_exceptions=True,
+                continue
-        )
+            try:
-        for r, content in zip(top, contents, strict=False):
+                content = await _fetch_full_article(url)
-            content_text = content if isinstance(content, str) else ""
+            except Exception:
+                content = None
             web_payload.append({
-                "url": r["url"],
+                "url": url,
-                "title": r.get("title", r["url"]),
+                "title": r.get("title", url),
                 "snippet": r.get("snippet", ""),
-                "content": content_text[:4000],
+                "content": (content or "")[:4000],
             })
 
     if not wiki_payload and not web_payload: