[v0.13.0][Feature] Add DeepSeek v4 initial support (#8648)

### What this PR does / why we need it?
**DeepSeek V4 Support**: Added support for DeepSeek V4 by introducing
new operators and infrastructure, including the Compressor operator and
associated tiling logic.

Note that: this PR is for v0.13.0. After this PR merge will release a
special release v0.13.0rc3. vLLM Ascend team will also work on main
branch rebase work soon.

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

### How was this patch tested?
- CI passed
- Do e2e test for
https://modelscope.cn/models/deepseek-ai/DeepSeek-V4-Flash

---------

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: GDzhu01 <809721801@qq.com>
Signed-off-by: LookAround0301 <lixushi@huawei.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
Signed-off-by: nwpu-zxr <zhouxuerong2@huawei.com>
Signed-off-by: WithHades<244036962@qq.com>
Signed-off-by: zhangsicheng5 <zhangsicheng5@huawei.com>
Signed-off-by: coder-fny <985619145@qq.com>
Signed-off-by: slippersss <slippersss@126.com>
Signed-off-by: yiz-liu  <liu_yizhou@outlook.com>
Signed-off-by: maoxx241 <maomaoyu870@gmail.com>
Signed-off-by: zhaozx-cn <zhaozx2116@163.com>
Signed-off-by: wxh571001500 <571001500@qq.com>
Signed-off-by: lcfenglinwan <lcfenglin@qq.com>
Signed-off-by: zhenwenqi_2024 <zhenwenqi_2022@qq.com>
Signed-off-by: anon189Ty <Stari_Falcon@outlook.com>
Signed-off-by: monologue815 <monologue815@qq.com>
Signed-off-by: Liexss <924834690@qq.com>
Signed-off-by: pinfa <1819563383@qq.com>
Signed-off-by: weinachuan<1173732899@qq.com>
Signed-off-by: chenchris2 <1349418798@qq.com>
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
Signed-off-by: wjunLu <wjunlu217@gmail.com>
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Co-authored-by: GDzhu01 <809721801@qq.com>

Author: wangxiyuan

.github/workflows/_e2e_test.yaml

@@ -84,6 +84,7 @@ jobs:
           VLLM_WORKER_MULTIPROC_METHOD: spawn
         if: ${{ inputs.type == 'light' }}
         run: |
+          . vllm_ascend/_cann_ops_custom/vendors/custom_transformer/bin/set_env.bash
           pytest -sv --durations=0 tests/e2e/singlecard/test_aclgraph_accuracy.py::test_piecewise_res_consistency
           pytest -sv --durations=0 tests/e2e/singlecard/test_quantization.py::test_qwen3_w8a8_quant
 
@@ -93,6 +94,7 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: max_split_size_mb:256
         if: ${{ inputs.type == 'full' }}
         run: |
+          . vllm_ascend/_cann_ops_custom/vendors/custom_transformer/bin/set_env.bash
           # We found that if running aclgraph tests in batch, it will cause AclmdlRICaptureBegin error. So we run
           # the test separately.
           # basic 
@@ -194,6 +196,7 @@ jobs:
           VLLM_WORKER_MULTIPROC_METHOD: spawn
         if: ${{ inputs.type == 'light' }}
         run: |
+          . vllm_ascend/_cann_ops_custom/vendors/custom_transformer/bin/set_env.bash
           pytest -sv --durations=0 tests/e2e/multicard/2-cards/test_qwen3_moe.py::test_qwen3_moe_distributed_mp_tp2_ep
 
       - name: Run vllm-project/vllm-ascend test (full)

.github/workflows/_unit_test.yaml

@@ -80,7 +80,12 @@ jobs:
             --ignore tests/ut/kv_connector/test_remote_decode_lifecycle.py \
             --ignore tests/ut/core/test_scheduler_dynamic_batch.py \
             --ignore tests/ut/kv_connector/test_mooncake_connector.py \
-            --ignore tests/ut/worker/test_worker_v1.py
+            --ignore tests/ut/worker/test_worker_v1.py \
+            --ignore tests/ut/distributed/test_parallel_state.py \
+            --ignore tests/ut/kv_connector/test_mooncake_layerwise_connector.py \
+            --ignore tests/ut/quantization/test_utils.py \
+            --ignore tests/ut/spec_decode/test_mtp_proposer.py \
+            --ignore tests/ut/test_platform.py
 
       - name: Upload coverage to Codecov
         # only upload coverage when commits merged

.gitignore

@@ -210,3 +210,5 @@ kernel_meta/
 # generated by CANN
 fusion_result.json
 csrc/output/
+csrc/build_out
+csrc/third_party

.pre-commit-config.yaml

@@ -4,7 +4,7 @@ default_install_hook_types:
 default_stages:
   - pre-commit # Run locally
   - manual # Run in CI
-exclude: 'examples/.*' # Exclude examples from all hooks by default
+exclude: 'examples/.*|csrc/.*' # Exclude examples from all hooks by default
 repos:
 - repo: https://github.com/codespell-project/codespell
   rev: v2.4.1