[Bugfix][Core] Fix CLI parallel flags losing to nested deploy parallel_config

[zwhzzz0821]

Purpose

Fix a precedence inversion introduced by the typed-config switch (#6849) that broke the NPU nightly HunyuanImage3 diffusion perf tests.

The legacy path StageConfig.to_omegaconf moves flat CLI parallel fields (tensor_parallel_size, ulysses_degree, ...) into the nested parallel_config dict via _apply_diffusion_parallel_runtime_overrides, so CLI flags win over a deploy YAML that nests its own parallel_config values. The new typed path _stage_engine_values skipped that step: CLI flags stayed flat, and _build_parallel_config then applied the nested deploy dict on top of the flat values — letting the deploy YAML beat the CLI.

In vllm_omni/deploy/hunyuan_image3_dit.yaml, the npu platform section sets nested parallel_config.tensor_parallel_size: 4, while the perf tests pass --tensor-parallel-size 2. The inversion turned the CLI's 2 back into the YAML's 4, so the stage demanded tp=4 × usp=2 = 8 devices on a 4-card NPU machine:

RuntimeError: Stage 0 requires 8 device(s) based on parallel_config, but 4 device(s) are available

This PR applies _apply_diffusion_parallel_runtime_overrides before reconcile_diffusion_attention_overrides in _stage_engine_values, mirroring the to_omegaconf order, and adds a regression test.

Fixes #7778

Test Plan

• New regression test test_diffusion_cli_parallel_overrides_beat_nested_deploy_parallel_config in tests/config/test_omni_config.py: deploy config with nested parallel_config.tensor_parallel_size=4 + CLI overrides tensor_parallel_size=2, ulysses_degree=2, asserting the resolved stage has tp=2, ulysses=2, world_size=4.
• Reproduced the failure on upstream main by resolving hunyuan_image3_dit.yaml with the perf test's CLI flags (platform forced to npu): before the fix the stage resolved to tp=4, usp=2, world_size=8; after the fix tp=2, usp=2, world_size=4.
• Full config/engine unit suites: pytest tests/config/ tests/engine/test_stage_engine_args.py tests/engine/test_async_omni_engine_stage_init.py.
• Real NPU e2e was not run locally (no NPU available); the NPU nightly lane should confirm.

vLLM Version: 0.29.0 (repo main)

vLLM-Omni Commit: 328d603

Test Result

• pytest tests/config/test_omni_config.py: 221 passed, 1 skipped
• pytest tests/config/ tests/engine/test_stage_engine_args.py tests/engine/test_async_omni_engine_stage_init.py: 884 passed, 2 skipped
• Reproduction script (platform forced to npu, --tensor-parallel-size 2 --ulysses-degree 2 against hunyuan_image3_dit.yaml):
  • Before: tensor_parallel_size=4, ulysses_degree=2, world_size=8 (matches the nightly failure)
  • After: tensor_parallel_size=2, ulysses_degree=2, sequence_parallel_size=2, world_size=4

[vllm-omni-review-bot]

This PR appears to belong to: docs/design/module/vllm_omni_config.md, docs/design/module/entrypoints.md, docs/design/module/stage_runtime.md.

Module owners: @alex-jw-brooks @lishunyang12 @NickCao

Routing: @alex-jw-brooks via module of the changed files, semantic router, CODEOWNERS; @lishunyang12 via module of the changed files, semantic router, CODEOWNERS; @NickCao via module of the changed files, CODEOWNERS

@zwhzzz0821, please review your own changes and leave a short self-review comment describing what you checked. PRs without author self-review may not be assigned a reviewer.

Please take a look when you have a chance. If you would like an automated review, mention @vllm-omni-review-bot in a comment.

[vllm-omni-review-bot]

Omni ReviewBot triage note

Automated triage of commit 328d603b6ec9 produced:

• Priority: high. Prompt maintainer attention is suggested.

These are automated triage suggestions only — the final decision belongs to the maintainers.

[NickCao]

LGTM, now it indeed "Mirror StageConfig.to_omegaconf".