[Feature] adapt to uva buffer and main2main (#6657)

### What this PR does / why we need it?
vllm model runner v2 use uva buffer to prepare input data, but npu
doesn't support uva yet, this pr implement a uvawrapper class to mimic
gpu's uva backend. what's more, this pr make some modifications to adapt
to the newer main branch.

### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
- vLLM main:
13397841ab

---------

Signed-off-by: Ronald1995 <ronaldautomobile@163.com>

vllm_ascend/worker/v2/states.py

@@ -17,11 +17,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-from contextlib import contextmanager
-
 import torch
-import vllm
-from vllm.v1.utils import CpuGpuBuffer
 from vllm.v1.worker.gpu.states import RequestState
 
 
@@ -36,7 +32,6 @@ class AscendRequestState(RequestState):
         num_speculative_steps: int,
         vocab_size: int,
         device: torch.device,
-        pin_memory: bool,
     ):
         super().__init__(
             max_num_reqs,
@@ -45,11 +40,7 @@ class AscendRequestState(RequestState):
             num_speculative_steps,
             vocab_size,
             device,
-            pin_memory,
         )
-        # because we will override these attribute, delete these attribute to
-        # make sure it's collected by python gc immediately.
-        del self.prefill_token_ids
         # vllm gpu_model_runner_v2 deprecate the seqs_lens_cpu attribute,
         # because they think most attention backends do not need it.
         # However, Ascend attention backend muse uses seqs_lens_cpu,
@@ -60,11 +51,6 @@ class AscendRequestState(RequestState):
             dtype=torch.int32,
             device="cpu",
         )
-        # NOTE(Ronald1995): Ascend NPUs do not support UVA yet,
-        # so we use CpuGpuBuffer to allocate prefill_token_ids buffer.
-        self.prefill_token_ids: CpuGpuBuffer = self._make_buffer(  # type: ignore
-            (self.max_num_reqs, self.max_model_len), dtype=torch.int32
-        )
 
     def add_request(
         self,
@@ -72,32 +58,12 @@ class AscendRequestState(RequestState):
         prompt_len,
         prefill_token_ids,
         num_computed_tokens,
-        sampling_params,
-        lora_request,
     ):
         super().add_request(
             req_id,
             prompt_len,
             prefill_token_ids,
             num_computed_tokens,
-            sampling_params,
-            lora_request,
         )
         req_idx = self.req_id_to_index[req_id]
         self.num_computed_tokens_cpu[req_idx] = num_computed_tokens
-
-
-@contextmanager
-def uva_wrapper():
-    """Context manager to disable UVA for Ascend NPUs."""
-
-    class UvaBufferWrapper:
-        def __init__(self, *args, **kwargs):
-            pass
-
-    try:
-        # TODO(Ronald1995): rectify this when NPU support uva.
-        vllm.v1.worker.gpu.states.UvaBuffer = UvaBufferWrapper
-        yield
-    finally:
-        pass