[feat] Introduce zero-copy to use YuanrongStorageClient for transmitting CPU Tensors in NPU environment

Co-authored-by: liwenlin<liwenlin8@huawei.com>



# message auto-generated for no-merge-commit merge:
!24 merge npu_zcopy into main

[feat] Introduce zero-copy to use `YuanrongStorageClient` for transmitting CPU Tensors in NPU environment

Created-by: Lexie-7
Commit-by: liwenlin
Merged-by: ascend-robot
Description: ### Summary

Transmitting CPU tensor in the NPU environment and achieving zero-copy.

### Change

Modified the `transfer_queue/storage/clients/yuanrong_client.py` , enable zero-copy transmission of CPU data  In the NPU branch.

See merge request: Ascend/TransferQueue!24

Author: ascend-robot

transfer_queue/storage/clients/yuanrong_client.py

@@ -260,7 +260,7 @@ def _batch_put(self, keys: list[str], values: list[Any]):
             for i in range(0, len(cpu_keys), CPU_DS_CLIENT_KEYS_LIMIT):
                 batch_keys = cpu_keys[i : i + CPU_DS_CLIENT_KEYS_LIMIT]
                 batch_values = cpu_values[i : i + CPU_DS_CLIENT_KEYS_LIMIT]
-                self._cpu_ds_client.mset(batch_keys, batch_values)
+                self.mset_zcopy(batch_keys, batch_values)
 
         else:
             #  All data goes through CPU path
@@ -356,9 +356,9 @@ def _batch_get(self, keys: list[str], shapes: list, dtypes: list) -> list[Any]:
             for i in range(0, len(cpu_keys), CPU_DS_CLIENT_KEYS_LIMIT):
                 batch_keys = cpu_keys[i : i + CPU_DS_CLIENT_KEYS_LIMIT]
                 batch_indices = cpu_indices[i : i + CPU_DS_CLIENT_KEYS_LIMIT]
-                raw_values = self._cpu_ds_client.get(batch_keys)
-                for idx, raw_val in zip(batch_indices, raw_values, strict=False):
-                    results[idx] = pickle.loads(raw_val)
+                objects = self.mget_zcopy(batch_keys)
+                for idx, obj in zip(batch_indices, objects, strict=False):
+                    results[idx] = obj
 
             return results