[Main] Upgrade PTA to 2.9.0

[wjunLu]

What this PR does / why we need it?

Upgrade PTA to 2.9.0

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.13.0
• vLLM main: vllm-project/vllm@d682094

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request upgrades the PyTorch Ascend (PTA) dependency to version 2.9.0. The changes consistently update the version number across various configuration files and documentation. My review identified a couple of outdated documentation links that were updated in this PR but still point to old versions. Correcting these will improve the user experience. Additionally, I noticed another outdated link in docs/source/installation.md that was not part of this PR's changes; you may want to address that as well for consistency.

[gemini-code-assist]

The link to the Ascend HDK version release notes is outdated. It still points to the documentation for CANN 8.3.RC2, but the required version has been updated to 8.5.0. This could cause confusion for users trying to set up the correct environment. Please update the link to point to the release notes for CANN 8.5.0.

Suggested change

	- CANN == 8.5.0 (Ascend HDK version refers to [here](https://www.hiascend.com/document/detail/zh/canncommercial/83RC2/releasenote/releasenote_0000.html))
	- CANN == 8.5.0 (Ascend HDK version refers to [here](https://www.hiascend.com/document/detail/zh/canncommercial/850/releasenote/releasenote_0000.html))

[gemini-code-assist]

The link to the Ascend HDK version release notes is outdated. It still points to the documentation for CANN 8.3.RC2, but the required version has been updated to 8.5.0. This could cause confusion for users. Please update the link to point to the release notes for CANN 8.5.0.

Suggested change

	- CANN == 8.5.0 (Ascend HDK 版本参考[这里](https://www.hiascend.com/document/detail/zh/canncommercial/83RC2/releasenote/releasenote_0000.html))
	- CANN == 8.5.0 (Ascend HDK 版本参考[这里](https://www.hiascend.com/document/detail/zh/canncommercial/850/releasenote/releasenote_0000.html))

[wjunLu]

• FAILED tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case1] - AssertionError: Test0:

E           AssertionError: Test0:
E           output_origin:	'\n\nSelect an assignment template'
E           outputs_gen:	"\n\nYour answer seems reasonable. Find out if you're right!\n\nSign up to access problem solutions.\n\nThat seems reasonable. Find out"

tests/e2e/model_utils.py:51: AssertionError

[wangxiyuan]

• test_npugraph_ex_res_consistency

I think we can update this test together

[wjunLu]

2/4-cards are all passed: https://github.com/vllm-project/vllm-ascend/actions/runs/21235020255/job/61107478634?pr=6112

[wjunLu]

• test_npugraph_ex_res_consistency

OK, I'll update this

[wjunLu]

• test_npugraph_ex_res_consistency

I think we can update this test together

I have to skip this case first since the outputs are not stable, see

2nd running result is:

============================== slowest durations ===============================
113.63s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_piecewise_res_consistency[cur_case1]
97.59s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_piecewise_res_consistency[cur_case0]
78.75s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case1]
70.35s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case0]
64.76s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_full_decode_only_res_consistency[cur_case1]
61.77s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_full_decode_only_res_consistency[cur_case0]

(12 durations < 0.005s hidden.  Use -vv to show these durations.)
=========================== short test summary info ============================
FAILED tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case1] - AssertionError: Test1:
output_origin:	"\n\nYour answer seems reasonable. Find out if you're right!\n\nSign up to access problem solutions.\n\nThat seems reasonable. Find out"
outputs_gen:	"\n\nI'm not sure how to approach this problem. I'm not sure if I should use the law of total probability or if I should use"
============= 1 failed, 5 passed, 2 warnings in 487.13s (0:08:07) ==============

1st running result is:

============================== slowest durations ===============================
120.80s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_piecewise_res_consistency[cur_case0]
105.53s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_piecewise_res_consistency[cur_case1]
86.36s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case1]
70.51s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case0]
69.33s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_full_decode_only_res_consistency[cur_case1]
67.09s call     tests/e2e/singlecard/test_aclgraph_accuracy.py::test_full_decode_only_res_consistency[cur_case0]

(12 durations < 0.005s hidden.  Use -vv to show these durations.)
=========================== short test summary info ============================
FAILED tests/e2e/singlecard/test_aclgraph_accuracy.py::test_npugraph_ex_res_consistency[cur_case1] - AssertionError: Test0:
output_origin:	'\n\nSelect an assignment template'
outputs_gen:	"\n\nYour answer seems reasonable. Find out if you're right!\n\nSign up to access problem solutions.\n\nThat seems reasonable. Find out"
============= 1 failed, 5 passed, 2 warnings in 519.88s (0:08:39) ==============

But I will find out if all 3 prompts should keep the same golden result or not.

[wjunLu]

• [local error] tests/e2e/singlecard/spec_decode/test_v1_spec_decode.py

======================================================== warnings summary ========================================================
<frozen importlib._bootstrap>:241
  <frozen importlib._bootstrap>:241: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute

<frozen importlib._bootstrap>:241
  <frozen importlib._bootstrap>:241: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
==================================================== short test summary info =====================================================
FAILED tests/e2e/singlecard/spec_decode/test_v1_spec_decode.py::test_llama_qwen_eagle_acceptance[True-False-None-3-eagle] - assert False

Adding requests: 100%|██████████████████████████████████████████████████████████████████████████████| 4/4 [00:00<00:00, 38.08it/s]
Processed prompts:   0%|                                | 0/4 [00:00<?, ?it/s, est. speed input: 0.00 toks/s, output: 0.00 toks/s](Worker pid=18518) INFO 01-22 09:20:59 [acl_graph.py:188] Replaying aclgraph
Processed prompts: 100%|██████████████████████| 4/4 [00:04<00:00,  1.25s/it, est. speed input: 33.32 toks/s, output: 62.23 toks/s]
(Worker pid=18518) INFO 01-22 09:21:04 [multiproc_executor.py:707] Parent process exited, terminating worker
acceptance_per_pos: [0.7205882352941176, 0.375, 0.17647058823529413]
golden: [0.74, 0.44, 0.29]
FAILED