[Pooling] MRV2 pooling shutdown model ref (#57737)

Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>

vllm/v1/worker/gpu/model_runner.py

@@ -2245,6 +2245,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
             self.aux_output_connector.close()
         self.cudagraph_manager = None
         self.fast_prefill = None
+        self.pooling_runner = None
         if hasattr(self, "kv_caches"):
             self.kv_caches.clear()
         if hasattr(self, "attn_groups"):