[main][refactor] Refactoring forward_context and model_runner_v1 (#1979)

### What this PR does / why we need it?

A refactoring of forward_context and model_runner_v1, add some context
which is necessary in model inference into forward_context, and refactor
dummy_run logic, make it more reasonable.
Some details for this PR:

Add `ascend_forward_context`;
Update mc2_v2 op, and support `active_mask` param;
Update scripts in examples dir;
refactor `dummy_run` logic;
Add soc_version for A2 and A3;

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

No change at user-facing.

### How was this patch tested?


- vLLM version: v0.10.0
- vLLM main:
57c22e57f9

Signed-off-by: zzzzwwjj <1183291235@qq.com>

vllm_ascend/models/deepseek_v2.py

@@ -377,20 +377,14 @@ class CustomDeepseekV2MoE(nn.Module):
                 attn_metadata: Optional[AttentionMetadata] = None,
                 replace_allreduce: bool = False) -> torch.Tensor:
 
-        if attn_metadata is None:
-            attn_metadata = get_forward_context().attn_metadata
+        forward_context = get_forward_context()
         # when profile runs, force experts to load balanced tokens
         # to avoid high memory consumption on a single rank.
-        # TODO: need a better flag to indicate whether in profile run or not.
-        if attn_metadata is None:
-            # for profile run
-            is_prefill = True
-            enable_force_load_balance = True
-        else:
-            is_prefill = attn_metadata.num_prefills > 0
-            enable_force_load_balance = False
-            if hasattr(attn_metadata, 'with_prefill_across_dp'):
-                is_prefill = is_prefill or attn_metadata.with_prefill_across_dp
+
+        enable_force_load_balance = forward_context.in_profile_run
+
+        is_prefill = forward_context.with_prefill
+
         # If this node is kv_consumer, we force the moe always runs in decode path to make sure
         # the behaviour aligned between dummy_run and normal model_execute.
         if self.kv_consumer:
@@ -572,9 +566,10 @@ class CustomDeepseekV2MLAAttention(DeepseekV2MLAAttention):
             hidden_states: torch.Tensor,
             kv_cache: Optional[torch.Tensor] = None,
             attn_metadata: Optional[AttentionMetadata] = None) -> torch.Tensor:
+        forward_context = get_forward_context()
         enable_multistream_mla = (self.enable_multistream_mla
                                   and attn_metadata is not None
-                                  and not attn_metadata.with_prefill_across_dp
+                                  and not forward_context.with_prefill
                                   and attn_metadata.num_decodes > 0)
         forward_kwargs = {"enable_multistream_mla": enable_multistream_mla}
         if self.q_lora_rank is not None: