[refactor] replace scattered business kwargs with typed request objects and explicit stage boundaries (#7024)

### What this PR does / why we need it?
Refactor `vllm_ascend/ops/fused_moe` to replace scattered MoE business
`**kwargs` with typed request objects and explicit stage boundaries.

- Prepare, dispatch, MLP, and quant stages now have clearer ownership.
- Main MoE path no longer depends on business `kwargs.get(...)` lookups.
- Comm and dispatcher interfaces are request-only on the main path.
- UTs can assert stage-level fields directly instead of inferring
behavior indirectly.

### Does this PR introduce _any_ user-facing change?
No.

### How was this patch tested?
CI passed.

---------

Signed-off-by: linfeng-yuan <1102311262@qq.com>

vllm_ascend/quantization/methods/base.py

@@ -18,19 +18,11 @@
 
 from abc import ABC, abstractmethod
 from collections.abc import Callable
-from enum import Enum
 from typing import Any
 
 import torch
 
-
-class QuantType(Enum):
-    """Quantization type enum for MoE schemes."""
-
-    NONE = 0
-    W8A8 = 1
-    W4A8 = 2
-    MXFP8 = 3
+from vllm_ascend.quantization.quant_type import QuantType
 
 
 class AscendLinearScheme(ABC):
@@ -245,7 +237,10 @@ class AscendMoEScheme(ABC):
         enable_force_load_balance: bool = False,
         log2phy: torch.Tensor | None = None,
         global_redundant_expert_num: int = 0,
-        **kwargs,
+        pertoken_scale: Any | None = None,
+        activation: str = "silu",
+        apply_router_weight_on_input: bool = False,
+        mc2_mask: torch.Tensor | None = None,
     ) -> torch.Tensor:
         """Forward computation for MoE layer.
 
@@ -268,7 +263,10 @@ class AscendMoEScheme(ABC):
             enable_force_load_balance: Whether to force load balancing.
             log2phy: Logical to physical expert mapping.
             global_redundant_expert_num: Number of redundant experts.
-            **kwargs: Additional keyword arguments.
+            pertoken_scale: Optional per-token activation scale from prepare stage.
+            activation: Expert MLP activation type.
+            apply_router_weight_on_input: Whether to pre-scale hidden states by router weights.
+            mc2_mask: Optional mask used by MC2 dispatch.
 
         Returns:
             Output tensor after MoE computation.