perf: concurrent fetches for multi-instance alert backends

Add _fetch_many() helper using ThreadPoolExecutor to query instances
in parallel. Refactors PeerTube, Mastodon, Lemmy, and SearXNG from
sequential to concurrent fetches. Also adds retries parameter to
derp.http.urlopen; multi-instance backends use retries=1 since
instance redundancy already provides resilience.

Worst-case wall time per backend drops from N*timeout to 1*timeout.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/derp/http.py

@@ -52,24 +52,25 @@ class _ProxyHandler(SocksiPyHandler, urllib.request.HTTPSHandler):
         return self.do_open(build, req)
 
 
-def urlopen(req, *, timeout=None, context=None):
+def urlopen(req, *, timeout=None, context=None, retries=None):
     """Proxy-aware drop-in for urllib.request.urlopen.
 
     Retries on transient SSL/connection errors with exponential backoff.
     """
+    max_retries = retries if retries is not None else _MAX_RETRIES
     opener = _get_opener(context)
     kwargs = {}
     if timeout is not None:
         kwargs["timeout"] = timeout
-    for attempt in range(_MAX_RETRIES):
+    for attempt in range(max_retries):
         try:
             return opener.open(req, **kwargs)
         except _RETRY_ERRORS as exc:
-            if attempt + 1 >= _MAX_RETRIES:
+            if attempt + 1 >= max_retries:
                 raise
             delay = 2 ** attempt
             _log.debug("urlopen retry %d/%d after %s: %s",
-                       attempt + 1, _MAX_RETRIES, type(exc).__name__, exc)
+                       attempt + 1, max_retries, type(exc).__name__, exc)
             time.sleep(delay)