[refactor] refactor weight trans nz and transpose (#4878)

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

Now `VLLM_ASCEND_ENABLE_NZ` will have three options:
0: disable nz;
1: only quant case enable nz;
2: enable nz as long as possible;

And `VLLM_ASCEND_ENABLE_NZ`=1 by default.

All cases are shown in the table below:

|  | W4A4 | W4A8 | W8A8 | fp16/bf16 | fp32 |
|---|---|---|---|---|---|
| trans nz | can't support nz | trans nz by default | trans nz by
default | trans nz when VLLM_ASCEND_ENABLE_NZ is 2 | can't support nz |
| transpose | only support not transpose case | only support transpose
case | only support transpose case | linear: only support not transpose
case<br>gmm: only support transpose case | same to fp16/bf16 |

Some exceptional cases:
1. MLAPO op need to do some additional processing on the weights,
including trans nz. If use MLAPO op, some weight will be transformed to
nz forcely;
2. MLA/SFA's weight `W_UV` will be used by op
`torch.ops._C_ascend.batch_matmul_transpose`, and this op can't support
nz currently;

### Does this PR introduce _any_ user-facing change?
Now fp16/bf16 weight will not trans nz by default.

### How was this patch tested?

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

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

vllm_ascend/utils.py

@@ -122,8 +122,22 @@ def _unregister_print_streams_on_exit():
 atexit.register(_unregister_print_streams_on_exit)
 
 
-def is_enable_nz():
-    return envs_ascend.VLLM_ASCEND_ENABLE_NZ
+def maybe_trans_nz(weight: torch.Tensor):
+    if not envs_ascend.VLLM_ASCEND_ENABLE_NZ:
+        # NZ is not enabled
+        return weight
+    if weight.dtype == torch.float:
+        # fp32 can not support NZ
+        return weight
+    elif weight.dtype in {torch.bfloat16, torch.float16}:
+        # bf16/fp16 will trans nz when VLLM_ASCEND_ENABLE_NZ is 2
+        if envs_ascend.VLLM_ASCEND_ENABLE_NZ == 2:
+            return torch_npu.npu_format_cast(weight, ACL_FORMAT_FRACTAL_NZ)
+        else:
+            return weight
+    else:
+        # quant weight will trans nz by default
+        return torch_npu.npu_format_cast(weight, ACL_FORMAT_FRACTAL_NZ)
 
 
 def _round_up(x: int, align: int):