[Feat] Unquantized Linear to nz and control all nz-cast (#3356)

### What this PR does / why we need it?
Currently, when executing to the Linear layer of models in vLLM-Ascend,
the weights format is ND in unquantized case and skipped ascend case.
This PR supplements the execution logic for Linear layer. We use a new
global variable: VLLM_ASCEND_ENABLE_NZ. When VLLM_ASCEND_ENABLE_NZ=1 and
CANN version is 8.3, the weights of the Linear layer will be converted
to FRACTAL_NZ, in both unquantized case and skipped ascend case. We also
use VLLM_ASCEND_ENABLE_NZ to control the existing NZ conversion, such as
w8a8-quantized case.

### Does this PR introduce _any_ user-facing change?
Add a new global variable VLLM_ASCEND_ENABLE_NZ. If you want to use NZ
format, you should set VLLM_ASCEND_ENABLE_NZ=1.

### How was this patch tested?

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

Signed-off-by: anon189Ty <Stari_Falcon@outlook.com>

vllm_ascend/torchair/torchair_sfa.py

@@ -24,6 +24,7 @@ from vllm_ascend.attention.utils import (AscendCommonAttentionMetadata,
 from vllm_ascend.multistream.base import MSAttentionMetadataSplitConfig
 from vllm_ascend.multistream.ms_split import model_input_split_v1_mla_attn
 from vllm_ascend.torchair.utils import TorchairCommonAttentionMetadata
+from vllm_ascend.utils import is_enable_nz
 from vllm_ascend.worker.npu_input_batch import InputBatch
 
 if TYPE_CHECKING:
@@ -841,7 +842,8 @@ class AscendSFATorchairImpl(MLAAttentionImpl):
         wd_qkv = wd_qkv.t().contiguous()
         wd_qkv = transdata(wd_qkv,
                            block_size=(16, 32)).unsqueeze(0).contiguous()
-        self.wd_qkv = torch_npu.npu_format_cast(wd_qkv, 29)
+        if is_enable_nz():
+            self.wd_qkv = torch_npu.npu_format_cast(wd_qkv, 29)
 
         kv_a_proj_deq_scl = self.kv_a_proj_with_mqa.deq_scale.clone()
         kv_a_proj_deq_scl = kv_a_proj_deq_scl.reshape(
@@ -874,7 +876,8 @@ class AscendSFATorchairImpl(MLAAttentionImpl):
             self.num_heads * (self.qk_nope_head_dim + self.qk_rope_head_dim),
             -1)
         wu_q = transdata(wu_q, block_size=(16, 32)).unsqueeze(0).contiguous()
-        self.wu_q = torch_npu.npu_format_cast(wu_q, 29)
+        if is_enable_nz():
+            self.wu_q = torch_npu.npu_format_cast(wu_q, 29)
 
         qb_deq_scl = self.q_proj.deq_scale.data.clone()
         qb_deq_scl = qb_deq_scl.reshape(