[fix] google: improve CAPTCHA detection (#5922)

- Detect HTTP 302 responses (Google redirecting to /sorry/index
  without the HTTP client following the redirect)
- Detect short HTML responses (<2000 bytes) containing "/sorry/"
  links (meta-refresh or JS redirect variants)

Instances with rotating IPs can set the `suspended_times.SearxEngineCaptcha` to
0 in the search settings [1], the next request will typically use a different
outgoing IP when rotating proxies are configured

[1] https://docs.searxng.org/admin/settings/settings_search.html

searx/engines/google.py

@@ -278,10 +278,28 @@ def get_google_info(params: "OnlineParams", eng_traits: EngineTraits) -> dict[st
     return ret_val
 
 
-def detect_google_sorry(resp):
+def detect_google_sorry(resp: "SXNG_Response"):
+    """Detect Google's bot-protection responses (CAPTCHA / sorry pages).
+
+    Google may block requests in several ways:
+
+    1. Redirect to sorry.google.com (standard CAPTCHA).
+    2. HTTP 302 redirect to ``/sorry/index?...`` on the same host -- when the
+       HTTP client doesn't follow the redirect, the response body is a short
+       HTML stub with a link to the sorry page.
+    3. Short HTML response (<2000 bytes) containing "/sorry/" -- a meta-refresh
+       or JS redirect variant.
+    """
+
     if resp.url.host == "sorry.google.com" or resp.url.path.startswith("/sorry"):
         raise SearxEngineCaptchaException()
 
+    if resp.status_code == 302:
+        raise SearxEngineCaptchaException()
+
+    if len(resp.text) < 2000 and "/sorry/" in resp.text:
+        raise SearxEngineCaptchaException()
+
 
 def request(query: str, params: "OnlineParams") -> None:
     """Google search request"""