Perf: 9 Performance-Fixes — SW by-v1072

Backend:
- DB: 3 neue Indizes (forum_posts thread+user, routes user) — Forum/Routen-Queries
- Caching: cache.py (TTL-Cache ohne neue Dependency) für 5 statische Listen
  (training_exercises, pflege_tipps, wiki_stats, wiki_gruppen, help_articles)
- diary.py + breeder_photos.py: Bildverarbeitung (ffmpeg/PIL/EXIF) per
  run_in_executor → blockiert Event-Loop nicht mehr
- scheduler.py: 11 kollidierende Jobs auf 5-Min-Intervalle gestaggert, coalesce=True
- social.py: ORDER BY RANDOM() ohne LIMIT in 2 Stellen gefixt
- alerts.py: Haversine-Loop bekommt SQL-Bounding-Box-Vorfilter

Frontend:
- sw.js: Tile-Cache mit LRU-Eviction (max 500 Einträge)
- admin.js: Event-Listener-Leak — Tab-Klicks per Delegation statt N Listener
- api.js: compressImage() Helper — Client-seitiges Resize auf max 2000px
  (HEIC/Videos/<500KB unverändert), integriert in 8 Upload-Stellen
  (diary, dog-profile×2, walks, poison, lost, health×2)

Bump APP_VER 1071 → 1072 (sw.js, app.js, main.py, index.html)

backend/routes/alerts.py

@@ -9,6 +9,7 @@ from auth import get_current_user_optional as get_optional_user
 router = APIRouter()
 
 _RADIUS_M = 20_000  # 20 km
+_RADIUS_KM = _RADIUS_M / 1000.0
 
 
 def _haversine(lat1: float, lon1: float, lat2: float, lon2: float) -> float:
@@ -20,15 +21,36 @@ def _haversine(lat1: float, lon1: float, lat2: float, lon2: float) -> float:
     return 2 * R * math.asin(math.sqrt(a))
 
 
+def _bbox(lat: float, lon: float, radius_km: float) -> tuple[float, float, float, float]:
+    """Bounding-Box-Approximation für lat/lon innerhalb radius_km."""
+    lat_delta = radius_km / 111.0
+    # cos darf bei Polen nicht 0 werden → mit kleinem Minimum absichern
+    cos_lat = max(abs(math.cos(math.radians(lat))), 0.01)
+    lon_delta = radius_km / (111.0 * cos_lat)
+    return (lat - lat_delta, lat + lat_delta, lon - lon_delta, lon + lon_delta)
+
+
 @router.get("")
 async def nearby_alerts(lat: float, lon: float, user=Depends(get_optional_user)):
     now = datetime.utcnow().isoformat()
+    lat_min, lat_max, lon_min, lon_max = _bbox(lat, lon, _RADIUS_KM)
     with db() as conn:
+        # Bounding-Box-Vorfilter per SQL (billig) → reduziert die Kandidaten
+        # auf ~10 Einträge statt "alle". Die exakte Haversine-Prüfung passiert
+        # anschließend in Python.
         poisons = conn.execute(
-            "SELECT lat, lon FROM poison WHERE geloest=0 AND expires_at > ?", (now,)
+            """SELECT lat, lon FROM poison
+               WHERE geloest=0 AND expires_at > ?
+                 AND lat BETWEEN ? AND ?
+                 AND lon BETWEEN ? AND ?""",
+            (now, lat_min, lat_max, lon_min, lon_max)
         ).fetchall()
         lost = conn.execute(
-            "SELECT lat, lon FROM lost_dogs WHERE is_active=1"
+            """SELECT lat, lon FROM lost_dogs
+               WHERE is_active=1
+                 AND lat BETWEEN ? AND ?
+                 AND lon BETWEEN ? AND ?""",
+            (lat_min, lat_max, lon_min, lon_max)
         ).fetchall()
         # Letzten Standort des Users für geo-basierte Push-Filter speichern
         if user: