Collect news text and image references without file attachments

Skip unsupported document formats and retire attachment work from active gap reporting. Preserve HTTP failures, normalize dynamic page content, and keep legacy bill queues resumable.

backend/src/alphapulse/core/public_http.py

@@ -19,6 +19,9 @@ async def public_address(url):
 async def fetch_public_response(url):
     async with httpx.AsyncClient(timeout=20, trust_env=False) as client:
         for _ in range(4):
+            from alphapulse.crawler.content_policy import file_url
+            if file_url(url):
+                raise ValueError('订阅地址不支持 PDF 或附件')
             parsed, address = await public_address(url)
             # Pin the checked address while retaining TLS hostname verification.
             target = httpx.URL(url).copy_with(host=address)
@@ -27,11 +30,18 @@ async def fetch_public_response(url):
                     location = response.headers.get('location')
                     if not location: raise ValueError('重定向缺少地址')
                     url = str(httpx.URL(url).join(location)); continue
-                chunks, size = [], 0
+                from scrapy.http import Headers
+                from alphapulse.crawler.content_policy import binary_headers
+                if binary_headers(Headers(response.headers.multi_items())):
+                    raise ValueError('订阅返回不支持的文件类型')
+                chunks, size, prefix = [], 0, b''
                 async for chunk in response.aiter_bytes():
                     size += len(chunk)
                     if size > 2*1024*1024: raise ValueError('响应超过2MiB限制')
                     chunks.append(chunk)
+                    prefix = (prefix + chunk)[:8]
+                    if prefix.lstrip().startswith(b'%PDF'):
+                        raise ValueError('订阅返回 PDF 文件')
                 body = b''.join(chunks)
                 # Bytes are already decompressed; Scrapy must not decode them again.
                 headers = [(k, v) for k, v in response.headers.multi_items()