[CI] Change Bagel online test environment variable VLLM_TEST_CLEAN_GPU_MEMORY to 0

[princepride]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Test Plan

Test Result

---

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. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (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)

[hsliuustc0106]

add comparison w/o changes

[princepride]

Previous

2 passed, 17 warnings in 553.85s (0:09:13)

GPU Memory Timeout warning in logs:

Pre-test cleanup note: [GPU Memory Timeout] Devices 0, 1, 2, 3, 4, 5, 6, 7 still don't meet memory condition after 124.9 seconds

After

2 passed, 17 warnings in 327.80s (0:05:27)

No timeout warnings. Cleanup correctly skipped:

GPU cleanup disabled

Why this change

The VLLM_TEST_CLEAN_GPU_MEMORY env var controls the autouse clean_gpu_memory_between_tests fixture in tests/conftest.py. When set to "1", it triggers wait_for_gpu_memory_to_clear() before every test case, polling all GPUs and waiting up to 125 seconds for memory usage to drop below 5%.

Problem: For online serving tests, the server process (omni_server fixture) is module-scoped — it stays alive across tests and holds GPU memory intentionally. The cleanup fixture cannot free this memory, so it always hits the 125s timeout before every test, adding ~2 minutes of pure waiting per test.

Why "0" is correct: All other online serving tests (test_qwen3_tts.py, test_qwen3_omni.py, test_video_generation_api.py, test_voxtral_tts.py, etc. — 7 files total) already set VLLM_TEST_CLEAN_GPU_MEMORY=0 for exactly this reason. The Bagel online test was the only one that had it set to "1", which was inconsistent.

Impact: ~226 seconds saved per run (553.85s -> 327.80s), and eliminates two misleading GPU Memory Timeout warnings from the test output. The GPU cleanup is still performed by OmniServer.__init__ (with enable_force=True) during server startup, so memory is properly managed when it actually matters.