[Doc] Fix Qwen3-Omni modality documentation examples

[psv666]

Fix incorrect modality descriptions and examples in the Qwen3-Omni online serving documentation.

Purpose

The Qwen3-Omni online serving docs contained several inaccuracies in the output-modality section that would mislead users into getting unexpected (or empty) responses:

1. Modality table mislabeled ["audio"] as "Text + Audio" — ["audio"] actually produces audio-only output. "Text + Audio" requires explicitly passing ["text", "audio"]. Corrected the table row.

2. The "Text + Audio" curl/Python examples used modalities=["audio"] — contradicting the section title. When ["text", "audio"] is requested the server returns two choices (index 0 = text, index 1 = audio), but the examples only decoded a single choice and used the wrong index. Updated the examples to pass ["text", "audio"], capture the full response, and extract text from choices[0] and audio from choices[1] (base64-decoded into output.wav).

3. The speaker (audio-only) Python example referenced choices[1] — audio-only responses contain a single choice, so the correct index is choices[0].

These fixes apply consistently to both docs/user_guide/examples/online_serving/qwen3_omni.md and examples/online_serving/qwen3_omni/README.md.

Test Plan

• Reviewed the rendered markdown diff for both files; both files make the same changes in the same sections.
• Verified the modality table now matches the actual server behavior: ["text"] → text only, ["audio"] → audio only, ["text","audio"] → text + audio.
• Verified the curl/Python snippets reference the correct choices indices for each modality combination.
• (Optional) Manual end-to-end verification: start the server with Qwen/Qwen3-Omni-30B-A3B-Instruct, run the corrected ["text","audio"] curl example, confirm two choices are returned and output.wav plays audio.

vLLM Version: 0.24.0

vLLM-Omni Commit: 7aa5c9a

Test Result

Documentation-only change; no runtime code modified. Verified by diff inspection that both copies of the example are consistent and the described choices indices match the server's actual response layout.

[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.

[amy-why-3459]

Thanks, LGTM.