fix(evidence): enforce active vector generations

harness/tht/vectorstore/rest_client.py

@@ -60,6 +60,7 @@ class VectorRestClient:
     def search_similar(
         self, table_name: str, query_embedding: list[float], limit_count: int,
         kinds: list[str] | None = None,
+        metadata_filter: dict | None = None,
     ) -> list[dict]:
         """Ricerca per similarità coseno su `vectors.<table_name>`: ritorna le righe
         `{id, similarity, metadata}` ordinate per similarity decrescente. Con `kinds`
@@ -72,6 +73,13 @@ class VectorRestClient:
             "limit_count": limit_count,
             "table_name": table_name,
         }
+        if metadata_filter is not None:
+            # ACTIVE corpus reads must never degrade to an unfiltered legacy RPC:
+            # filtering after LIMIT is incomplete and could expose stale generations.
+            return self._call(
+                "search_similar",
+                {**args, "kinds": kinds, "metadata_filter": metadata_filter},
+            ) or []
         if kinds is not None:
             try:
                 return self._call("search_similar", {**args, "kinds": kinds}) or []
@@ -118,3 +126,12 @@ class VectorRestClient:
                 return int(payload[0]["upserted"])
             return len(payload)
         return len(rows)
+
+    def delete_generation(self, table_name: str, generation: str) -> int:
+        payload = self._call(
+            "delete_vector_generation",
+            {"table_name": table_name, "kind": "evidence", "generation": generation},
+        )
+        if isinstance(payload, dict):
+            return int(payload.get("deleted", 0))
+        return 0