[Draft][WIP][Feature]cpu offload connector

[lidenghui1110]

What this PR does / why we need it?

This PR implements cpu offload connector to enable NPU kv cache offload to host DRAM.
This PR depend on vllm changes with starting a metadata-server process. Metadata-server manages cpu_kv_cache and offers rpc functions for the connector to call. It is designed to support shared-kv-cache between DP EngineCore.
Code of metadata-server is on working, we are trying to implement it in vllm-ascend to avoid long-term pull request merge in vllm.

Does this PR introduce any user-facing change?

user enable cpu offload with following params

 --kv-transfer-config \
    '{
    "kv_connector":"CPUOffloadingConnector",
        "kv_connector_module_path": "vllm_ascend.distributed.kv_transfer.cpu_offloading_connector",
        "kv_role":"kv_both", "kv_connector_extra_config": {"swap_in_threshold": 0, "cpu_swap_space_gb": 800}
    }'

How was this patch tested?

• vLLM version: v0.9.1
• vLLM main: vllm-project/vllm@110df74

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.