[Graph][Fusion]Add new pattern for AddRmsnormQuant with SP. (#5077)

### What this PR does / why we need it?
1. In addition to
[#4168](https://github.com/vllm-project/vllm-ascend/pull/4168),
[#5011](https://github.com/vllm-project/vllm-ascend/pull/5011)， this PR
adds two more pattern for AddRmsnormQuant with SP enabled. The key
difference is to insert an additional `maybe_all_gather_and_maybe_unpad`
between `addrmsnorm` and `quantize`.
2. This PR also introduce another api `torch.ops.vllm.quantize`, so that
we pass `input_scale` and `input_scale_reciprocal` at the same time.
This is because `npu_add_rms_norm_quant` and `npu_quantize` requires
different `div_mode`. To avoid introducing additional reciprocal
calculation in runtime, we have to pass both of them to quantize api.
3. Removes redundant `AscendQuantRmsnorm`.


- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: Angazenn <supperccell@163.com>

vllm_ascend/ops/linear_op.py

@@ -545,8 +545,7 @@ class SequenceRowParallelOp(CustomRowParallelOp):
         from vllm.model_executor.layers.linear import UnquantizedLinearMethod
 
         from vllm_ascend.quantization.quant_config import AscendLinearMethod
-        from vllm_ascend.quantization.w8a8 import (AscendW8A8LinearMethod,
-                                                   quant_per_tensor)
+        from vllm_ascend.quantization.w8a8 import AscendW8A8LinearMethod
 
         # For unquant
         if mmrs_fusion and isinstance(self.layer.quant_method,
@@ -568,8 +567,9 @@ class SequenceRowParallelOp(CustomRowParallelOp):
                 and isinstance(self.layer.quant_method.quant_method,
                                AscendW8A8LinearMethod)):
             if x.dtype != torch.int8:
-                x_quant = quant_per_tensor(
-                    x, self.layer.aclnn_input_scale_reciprocal,
+                x_quant = torch.ops.vllm.quantize(
+                    x, self.layer.aclnn_input_scale,
+                    self.layer.aclnn_input_scale_reciprocal,
                     self.layer.aclnn_input_offset)
             else:
                 x_quant = x