[Accuracy Benchmark]Add VBench video accuracy benchmarks t2v and i2v

[david6666666]

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

Purpose

Add video accuracy benchmark scaffolding for vLLM-Omni with:

• VBench runner for text_to_video
• VBench-I2V runner for image_to_video
• shared /v1/videos client and benchmark utilities
• partial-score summarization for L4 smoke coverage
• benchmark unit tests and L4 e2e smoke test entrypoints

Scope of this PR:

• add benchmarks/accuracy/common.py
• add benchmarks/accuracy/text_to_video/run_vbench.py
• add benchmarks/accuracy/image_to_video/run_vbench_i2v.py
• add benchmark utility tests
• add tests/e2e/accuracy/ smoke test scaffolding for VBench and VBench-I2V

Behavior notes:

• text_to_video uses VBench
• image_to_video uses VBench-I2V
• import is based on a repo checkout via --vbench-root / VBENCH_REPO_ROOT, not vendoring
• L4 smoke only reports partial benchmark summaries and does not claim official total benchmark scores
• this PR does not modify .buildkite/test-merge.yml or .buildkite/test-ready.yml

Test Plan

• Static validation:
  • python -m py_compile benchmarks/accuracy/common.py benchmarks/accuracy/text_to_video/run_vbench.py benchmarks/accuracy/image_to_video/run_vbench_i2v.py tests/benchmarks/test_video_accuracy_bench_utils.py tests/e2e/accuracy/conftest.py tests/e2e/accuracy/test_vbench_t2v_l4_smoke.py tests/e2e/accuracy/test_vbench_i2v_l4_smoke.py
• Targeted local assertions:
  • validate summarize_vbench_results
  • validate summarize_vbench_i2v_results
  • validate balanced subset selection
  • validate vbench_import_context prefers explicit repo checkout over preinstalled modules
• Remote validation to run after merge or in a proper benchmark environment:
  • pytest tests/benchmarks/test_video_accuracy_bench_utils.py
  • pytest tests/e2e/accuracy/test_vbench_t2v_l4_smoke.py
  • pytest tests/e2e/accuracy/test_vbench_i2v_l4_smoke.py

Test Result

• Passed:
  • python -m py_compile on all newly added benchmark and test files
  • direct Python validation for partial-summary logic, balanced selection logic, parser construction, and VBench repo import override behavior
• Not run locally:
  • pytest for the new tests, because the local environment fails during existing repo test bootstrap with ModuleNotFoundError: No module named 'soundfile' from tests/conftest.py
• Needs remote verification:
  • end-to-end VBench / VBench-I2V execution against a real video serving environment
  • L4 smoke runs with actual models and VBench repo checkout / dataset availability

---

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)