refactor(memory): consolidate memory search services and update model client handling

- Consolidate memory search services by removing separate content_search.py and perceptual_search.py
- Update model client handling in base_pipeline.py to use ModelApiKeyService for LLM client initialization
- Add new prompt files and modify existing services to support consolidated search architecture
- Refactor memory read pipeline and related services to use updated model client approach

api/app/repositories/neo4j/create_indexes.py

@@ -19,7 +19,8 @@ async def create_fulltext_indexes():
         # """)
         # 创建 Entities 索引
         await connector.execute_query("""
-            CREATE FULLTEXT INDEX entitiesFulltext IF NOT EXISTS FOR (e:ExtractedEntity) ON EACH [e.name]
+            CREATE FULLTEXT INDEX entitiesFulltext IF NOT EXISTS 
+            FOR (e:ExtractedEntity) ON EACH [e.name, e.description, e.aliases]
             OPTIONS { indexConfig: { `fulltext.analyzer`: 'cjk' } }
         """)
 
@@ -139,6 +140,16 @@ async def create_vector_indexes():
         await connector.close()
 
 
+async def create_user_indexes():
+    connector = Neo4jConnector()
+    await connector.execute_query(
+        """
+        CREATE INDEX user_perceptual IF NOT EXISTS
+        FOR (p:Perceptual) ON (p.end_user_id);
+        """
+    )
+
+
 async def create_unique_constraints():
     """Create uniqueness constraints for core node identifiers.
     Ensures concurrent MERGE operations remain safe and prevents duplicates.

api/app/repositories/neo4j/graph_search.py

@@ -45,14 +45,17 @@ def cosine_similarity_search(
     vectors: np.ndarray = np.array(vectors, dtype=np.float32)
     vectors_norm = vectors / np.linalg.norm(vectors, axis=1, keepdims=True)
     query: np.ndarray = np.array(query, dtype=np.float32)
-    query_norm = query / np.linalg.norm(query)
+    norm = np.linalg.norm(query)
+    if norm == 0:
+        return {}
+    query_norm = query / norm
 
     similarities = vectors_norm @ query_norm
     similarities = np.clip(similarities, 0, 1)
     top_k = min(limit, similarities.shape[0])
     if top_k <= 0:
         return {}
-    top_indices = np.argpartition(-similarities, top_k - 1)[-top_k:]
+    top_indices = np.argpartition(-similarities, top_k - 1)[:top_k]
     top_indices = top_indices[np.argsort(-similarities[top_indices])]
     result = {}
     for idx in top_indices:
@@ -510,7 +513,7 @@ async def search_graph_by_embedding(
     task_keys = []
 
     for node_type in include:
-        tasks.append(search_by_embedding(connector, node_type, end_user_id, embedding, limit))
+        tasks.append(search_by_embedding(connector, node_type, end_user_id, embedding, limit*2))
         task_keys.append(node_type.value)
 
     task_results = await asyncio.gather(*tasks, return_exceptions=True)