fix: disable all model preload to prevent CPU crash

entrypoint.sh

@@ -31,37 +31,8 @@ log "Ensuring cache table exists..."
 python hue_portal/manage.py createcachetable
 log "Cache table ready."
 
-log "Preloading all models to avoid first-request timeout..."
-
-python -c "
-import os
-import sys
-os.environ.setdefault('DJANGO_SETTINGS_MODULE', 'hue_portal.hue_portal.settings')
-import django
-django.setup()
-
-print('[ENTRYPOINT] 🔄 Starting model preload...', flush=True)
-
-# 1. Preload Embedding Model (BGE-M3)
-try:
-    print('[ENTRYPOINT] 📦 Preloading embedding model (BGE-M3)...', flush=True)
-    from hue_portal.core.embeddings import get_embedding_model
-    embedding_model = get_embedding_model()
-    if embedding_model:
-        print('[ENTRYPOINT] ✅ Embedding model preloaded successfully', flush=True)
-    else:
-        print('[ENTRYPOINT] ⚠️ Embedding model not loaded', flush=True)
-except Exception as e:
-    print(f'[ENTRYPOINT] ⚠️ Embedding model preload failed: {e}', flush=True)
-
-# 2. SKIP LLM Preload (CPU optimization - too heavy, will load on first request)
-print('[ENTRYPOINT] ⏭️ Skipping LLM preload (CPU optimization - will load lazily on first request)', flush=True)
-
-# 3. SKIP Reranker Preload (CPU optimization - too heavy, will load on first request)
-print('[ENTRYPOINT] ⏭️ Skipping reranker preload (CPU optimization - will load lazily on first request)', flush=True)
-
-print('[ENTRYPOINT] ✅ Model preload completed', flush=True)  # v2.0-preload-all
-" || log "⚠️ Model preload had errors (models will load on first request)"
+# Skip model preload to prevent CPU overload (models will load lazily on first request)
+log "⏭️ Skipping model preload (CPU optimization - models will load on first request)"
 
 log "Starting Gunicorn on port ${PORT:-7860}..."