fix: graceful fallback when colpali_engine not installed

colpali_retriever imports are now conditional so /compare endpoint
works on HF Spaces without colpali_engine; also fix local Qdrant path.

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

src/colpali_retriever.py

@@ -4,17 +4,22 @@ Model: vidore/colpali-v1.2 (multi-vector, MaxSim scoring)
 So sánh với pipeline text hiện tại để thấy sự khác biệt trên tài liệu có bảng biểu.
 """
 import os
-import torch
 from pathlib import Path
-from PIL import Image
-from colpali_engine.models import ColPali, ColPaliProcessor
 from qdrant_client import QdrantClient
 from qdrant_client.models import (
     Distance, VectorParams, MultiVectorConfig,
-    MultiVectorComparator, PointStruct, NamedSparseVector,
+    MultiVectorComparator, PointStruct,
 )
 from dotenv import load_dotenv
 
+try:
+    import torch
+    from PIL import Image
+    from colpali_engine.models import ColPali, ColPaliProcessor
+    COLPALI_AVAILABLE = True
+except ImportError:
+    COLPALI_AVAILABLE = False
+
 load_dotenv()
 
 COLPALI_MODEL = "vidore/colpali-v1.2"
@@ -26,6 +31,8 @@ _processor = None
 
 
 def _get_model():
+    if not COLPALI_AVAILABLE:
+        raise RuntimeError("colpali_engine not installed")
     global _model, _processor
     if _model is None:
         device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -41,11 +48,7 @@ def _get_model():
 
 
 def _get_client() -> QdrantClient:
-    url = os.getenv("QDRANT_URL")
-    api_key = os.getenv("QDRANT_API_KEY")
-    if url and api_key:
-        return QdrantClient(url=url, api_key=api_key, timeout=60)
-    return QdrantClient(path="qdrant_storage")
+    return QdrantClient(path="qdrant_colpali")
 
 
 def _get_or_create_collection(client: QdrantClient):