[Feature] Add PD separation feature (#432)

### What this PR does / why we need it?
Adapt Disaggregated Prefill feature onto Ascend device

### Does this PR introduce _any_ user-facing change?

no

### How was this patch tested?

The test usage has been provided alongwith the PR, in
examples/offline_disaggregated_prefill_npu.py
To run it, do this
```
export PROMPT_DEVICE_ID=0,1
export DECODE_DEVICE_ID=2,3
python examples/offline_disaggregated_prefill_npu.py
```

---------

Signed-off-by: ZihuiQian <qianzihui@huawei.com>
Co-authored-by: ZihuiQian <qianzihui@huawei.com>

vllm_ascend/worker/worker_v1.py

@@ -25,7 +25,8 @@ import torch.nn as nn
 import torch_npu
 from vllm import envs
 from vllm.config import VllmConfig
-from vllm.distributed import (ensure_model_parallel_initialized,
+from vllm.distributed import (ensure_kv_transfer_initialized,
+                              ensure_model_parallel_initialized,
                               init_distributed_environment,
                               set_custom_all_reduce)
 from vllm.logger import logger
@@ -197,6 +198,7 @@ class NPUWorker(WorkerBase):
         ensure_model_parallel_initialized(
             self.parallel_config.tensor_parallel_size,
             self.parallel_config.pipeline_parallel_size)
+        ensure_kv_transfer_initialized(self.vllm_config)
 
     def _init_profiler(self):
         # Torch profiler. Enabled and configured through env vars:
@@ -230,4 +232,4 @@ class NPUWorker(WorkerBase):
                 on_trace_ready=torch_npu.profiler.tensorboard_trace_handler(
                     torch_profiler_trace_dir))
         else:
-            return None
+            return None