[Refactor] Adapt deepseek-v3.2 to vllm 0.11.0 (#3432)

### What this PR does / why we need it?
Adapt deepseek-v3.2 to vllm 0.11.0, removing the useless patch.

The final goal is to remove all the patches and align the code arch to
vllm, thus we need to do the following work in next prs.
TODO:
- [x] remove patch on attention spec
- [ ] refactor the kvcache creation logic

### Does this PR introduce _any_ user-facing change?
N/A

### How was this patch tested?
1. CI passed with existing test.
2. Test pass with deepseek-v3.2-exp


- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: MengqingCao <cmq0113@163.com>

vllm_ascend/attention/sfa_v1.py

@@ -510,7 +510,6 @@ class AscendSFAImpl(MLAAttentionImpl):
 
         ascend_config = get_ascend_config()
         self.enable_shared_expert_dp = ascend_config.enable_shared_expert_dp
-        self.enable_prefetch = ascend_config.enable_prefetch
         self.enable_kv_nz = ascend_config.torchair_graph_config.enable_kv_nz
 
         vllm_config = get_current_vllm_config()
@@ -690,6 +689,8 @@ class AscendSFAImpl(MLAAttentionImpl):
             topk_indices = self.indexer_select(hidden_states_decode,
                                                decode_q_c,
                                                attn_metadata=attn_metadata,
+                                               cos=cos,
+                                               sin=sin,
                                                kv_cache=kv_cache)
 
             query_states = (decode_q_nope, decode_q_pe)
@@ -778,6 +779,8 @@ class AscendSFAImpl(MLAAttentionImpl):
             topk_indices = self.indexer_select(x=hidden_states_prefill,
                                                qr=prefill_qr,
                                                kv_cache=kv_cache,
+                                               cos=cos,
+                                               sin=sin,
                                                attn_metadata=attn_metadata)
             query_states = (prefill_q_nope, prefill_q_pe)
             key_states = (prefill_k_nope, prefill_k_pe)
@@ -920,17 +923,15 @@ class AscendSFAImpl(MLAAttentionImpl):
         x: torch.Tensor,
         qr: torch.Tensor,
         kv_cache: Tuple[torch.Tensor, torch.Tensor, torch.Tensor],
+        cos,
+        sin,
         attn_metadata: M,
     ):
         if attn_metadata.prefill is not None:
-            cos = attn_metadata.prefill.cos
-            sin = attn_metadata.prefill.sin
             actual_seq_lengths_query = attn_metadata.prefill.query_lens
             actual_seq_lengths_key = attn_metadata.prefill.seq_lens
             block_table = attn_metadata.prefill.block_table
         elif attn_metadata.decode is not None:
-            cos = attn_metadata.decode.cos
-            sin = attn_metadata.decode.sin
             actual_seq_lengths_query = attn_metadata.decode.actual_seq_lengths_q
             actual_seq_lengths_key = attn_metadata.decode.seq_lens
             block_table = attn_metadata.decode.block_table