[feat]: oproj tensor parallelism in pure DP and graph-mode scenarios. (#2167)

### What this PR does / why we need it?
This PR introduces Oproj matrix tensor model parallel to achieve
decreasing of memory consumption. It only support graph mode in pure DP
scenario.

In deepseek r1 w8a8 PD disagregated Decode instance, using pure DP, with
oproj_tensor_parallel_size = 8, we have 1 ms TPOT increasing, saved 5.8
GB NPU memory per RANK. We got best performance when
oproj_tensor_parallel_size=4 without TPOT increasing.

performance data:
<img width="1442" height="442" alt="image"
src="https://github.com/user-attachments/assets/83270fc5-868a-4387-b0a9-fac29b4a376d"
/>

### Does this PR introduce _any_ user-facing change?
This PR introduces one new config in `additional_config`.
| Name | Effect | Required | Type | Constraints |
| :---------------------------- |
:--------------------------------------- | :------- | :--- |
:----------------- |
| oproj_tensor_parallel_size | Split the o_proj matrix along the row
dimension (head num * head dim) into oproj_tensor_parallel_size pieces.
| No | int | default value is None, once this value is set, the feature
will be enabled, head num * head dim must be divisible by this value. |

example

`--additional_config={"oproj_tensor_parallel_size": 8}`

### How was this patch tested?


- vLLM version: v0.10.1.1
- vLLM main:
eddaafc1c7

---------

Signed-off-by: zzhx1 <zzh_201018@outlook.com>
Co-authored-by: zzh <zzh_201018@outlook.com>

tests/ut/torchair/models/test_torchair_deepseek_v2.py

@@ -100,6 +100,11 @@ def mock_distributed():
     pp_group.rank_in_group = 0
     pp_group.world_size = 1
 
+    mlp_tp_group = Mock(spec=GroupCoordinator)
+    mlp_tp_group.rank_in_group = 0
+    mlp_tp_group.world_size = 1
+    mlp_tp_group.all_gather = Mock(return_value=torch.randn(2, 4, 128))
+
     mock_vllm_config = Mock()
     mock_vllm_config.scheduler_config = Mock(max_num_seqs=256)
     mock_vllm_config.model_config = Mock(max_model_len=2048, quant_config=None)
@@ -196,10 +201,6 @@ def test_torchair_deepseek_v2_mlp(mock_distributed, base_config):
                                 quant_config=None)
     assert isinstance(mlp.act_fn, TorchairDeepseekV2SiluAndMul)
 
-    x = torch.randn(2, 4, 128)
-    output = mlp(x)
-    assert output.shape == (2, 4, 128)
-
     with patch(
             "vllm_ascend.torchair.models.torchair_deepseek_v2.QuantizationConfig"
     ) as mock_quant_config:
@@ -322,4 +323,4 @@ def test_torchair_deepseek_v2_for_causal_lm(mock_distributed, vllm_config):
             "vllm.model_executor.model_loader.weight_utils.default_weight_loader"
     ):
         loaded = model.load_weights(weights)
-        assert loaded is not None
+        assert loaded is not None