perf: seed alert seen IDs in background on add

Reply immediately with empty seen, run a silent _poll_once in a
background task to populate seen IDs, then start the poller.
Eliminates the 30-120s blocking wait from 27 sequential backend queries.

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

plugins/alert.py

@@ -2054,19 +2054,6 @@ async def cmd_alert(bot, message):
             await bot.reply(message, f"Alert limit reached ({_MAX_SUBS})")
             return
 
-        # Seed: query all backends to populate seen lists (prevents flood)
-        loop = asyncio.get_running_loop()
-        seen: dict[str, list[str]] = {}
-        for tag, backend in _BACKENDS.items():
-            try:
-                items = await loop.run_in_executor(None, backend, keyword)
-                ids = [item["id"] for item in items]
-                if len(ids) > _MAX_SEEN:
-                    ids = ids[-_MAX_SEEN:]
-                seen[tag] = ids
-            except Exception:
-                seen[tag] = []
-
         now = datetime.now(timezone.utc).isoformat()
         data = {
             "keyword": keyword,
@@ -2077,16 +2064,21 @@ async def cmd_alert(bot, message):
             "added_at": now,
             "last_poll": now,
             "last_error": "",
-            "seen": seen,
+            "seen": {},
         }
         _save(bot, key, data)
         _subscriptions[key] = data
-        _start_poller(bot, key)
 
-        counts = ", ".join(f"{len(seen.get(t, []))} {t}" for t in _BACKENDS)
+        # Seed seen IDs in background (silent poll), then start the poller
+        async def _seed():
+            await _poll_once(bot, key, announce=False)
+            _start_poller(bot, key)
+
+        asyncio.create_task(_seed())
+
         await bot.reply(
             message,
-            f"Alert '{name}' added for: {keyword} ({counts} existing)",
+            f"Alert '{name}' added for: {keyword} (seeding in background)",
         )
         return