[BugFix] Add field validators for /v1/audio/generate request

[Shaun-Walsh]

Summary

• Add @field_validator("input") rejecting empty/whitespace prompts
• Add gt=0 constraint to audio_length
• Add gt=0, le=1000 constraints to guidance_scale
• Add ge=1, le=1000 constraints to num_inference_steps
• Unskip all 8 issue [Bug]: Summary of Parameter Issues in Entrypoints Interface Exceptions #3649 tests for this endpoint
• Update audio_length_negative and num_inference_steps_negative test expectations to match entrypoint-level Pydantic validation (previously relied on engine-level checks)

Partial fix for #3649.

Test plan

• Verified all validators locally with Pydantic v2 (15/15 cases pass)
• Confirmed no file overlap with PRs [Bugfix] Fix audio.format parameter silently ignored in chat completions #4718/[Bugfix] Return error for unsupported output modalities instead of empty choices #4720
• CI passes on existing test suite
• Live validation with stable-audio-open-1.0 (requires L4)

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[Shaun-Walsh]

Hi — this PR is a dependency for the Red Hat Midstream build of vllm-omni. CI is green and it's ready for review. Could we get eyes on this when you have a chance? Happy to address any feedback quickly. Thanks!

[vllm-omni-review-bot]

Omni ReviewBot: no human activity for 48 days

@Shaun-Walsh this pull request has had no human commit, comment or review since 2026-07-21. Per repository policy it may be closed if it stays inactive.

To keep it moving, any one of these is enough: push an update, reply to the open blocker, or post the current plan and timeline.

[Shaun-Walsh]

Hi @yenuo26 & @NickCao @alex-jw-brooks
Rebased on main to resolve merge conflicts (was stale since 2026-07-21).

Two conflicts, both from #6598 ("Handle DoS Overflow Cases") having landed on main after this branch forked, which widened several numeric field bounds to full int64 range:

• vllm_omni/entrypoints/openai/protocol/audio.py: num_inference_steps bound — kept this PR's le=1000 (matches the num_inference_steps_above_max test added here) over main's le=_INT64_MAX.
• tests/dfx/reliability/invalid_param_test/test_invalid_audio_diffusion.py: num_inference_steps_zero/_negative expected-error-text — kept the type-slug style ("greater_than_equal") for consistency with the rest of the file over main's human-readable msg style ("greater than or equal to 1") from #6827. Cosmetic only — Pydantic v2 errors include both fields in the response body, so either matches the "all substrings present" assertion helper.

No functional changes beyond the rebase itself. Current plan/timeline unchanged from the PR description; still blocked on CI + live validation with stable-audio-open-1.0 (L4).