[CI] Refactor test_sequence_parallel.py and add a warmup run for more accurate performance stat

[mxuax]

Purpose

This PR refactors test_sequence_parallel.py, and adds a warmup run for more accurate performance stat to fix the issue6 mentioned in #1143. The redundant tests are also removed to speedup the test.

Issue 6: Test Validity Anomaly (Ring vs. Ulysses)
Evidence: In 4-GPU Sequence Parallel tests:
Ulysses: ~29.0s execution.
Ring/Hybrid: ~0.3s execution.
The reason is the time consumption during NCCL initialization. By adding a warm-up run, all those can be eliminated. 

Test Plan

pytest tests/e2e/offline_inference/test_sequence_parallel.py -v -s

Test Result

Hardware: H800
Time duration 388.15s (0:06:28)

======================================================================
SUMMARY
======================================================================
Mode            GPUs   Size       Baseline     SP           Speedup    Status
----------------------------------------------------------------------
ulysses-2       2      256x256   95ms         110ms        0.86x      PASS
ring-2          2      256x256   95ms         168ms        0.57x      PASS
hybrid-2x2      4      256x256   N/A          314ms        N/A        PASS
ulysses-4       4      272x272   N/A          9168ms        N/A        PASS
======================================================================

NOTE: The performance downgrade is because we use a very small image size(256*256), and the generation bottleneck is communication-bound instead of computational-bound.

---

Essential Elements of an Effective PR Description Checklist

• [ - ] The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• [ - ] The test plan, such as providing test command.
• [ - ] The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e483ab08fb

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[mxuax]

This PR is ready. @congw729 @wtomin @SamitHuang @ZJY0516 @hsliuustc0106

[congw729]

The codes are fine. Please rebase to origin/main branch to clean the commits log.

[hsliuustc0106]

this looks much better for each single function @yenuo26 @congw729

[yenuo26]

this looks much better for each single function @yenuo26 @congw729

Buildkite automatically collapses logs starting with "Running". If we want to achieve this effect, we can add a fixture function in conftest.py to add "Running" information before each test function runs.