[P/D] NixlConnector use cache device index for memory registration (#18969)

Signed-off-by: Piotr Tarasiewicz <ptarasiewicz@nvidia.com>

vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py

@@ -488,7 +488,8 @@ class NixlConnectorWorker:
             for cache in cache_list:
                 base_addr = cache.data_ptr()
                 region_len = self.num_blocks * self.block_len
-                caches_data.append((base_addr, region_len, self.rank, ""))
+                caches_data.append(
+                    (base_addr, region_len, cache.device.index, ""))
                 kv_caches_base_addr.append(base_addr)
         self.kv_caches_base_addr[self.engine_id] = kv_caches_base_addr
         self.num_regions = len(caches_data)