Fix stale _paged_kv_cache reference in test_paged_prefix_caching (#207)

Author: LxYuan0420

tests/test_paged_prefix_caching.py

@@ -26,7 +26,7 @@ def _make_paged_runner(num_layers: int = 2) -> mr.MetalModelRunner:
     runner = mr.MetalModelRunner.__new__(mr.MetalModelRunner)
     runner.model = MagicMock()
     runner._is_stt = False
-    runner._paged_kv_cache = MagicMock()  # non-None enables paged path
+    runner._paged_attention_backend = MagicMock()  # non-None enables paged path
     runner._paged_block_size = 4
     runner._paged_request_seq_lens = {}
     runner._request_states = {}