docs: clarify vLLM video upload and source indicator behavior

[shihanqu]

Summary

• add vLLM quick-start documentation notes for multimodal video usage
• clarify that No sources found can be expected for video-only turns without RAG citations
• clarify that File type video/mp4 is not supported for processing refers to file-processing/retrieval path, not necessarily model multimodal inference failure
• add a minimal direct API health check snippet for vLLM model listing

Why

Users connecting Open WebUI to vLLM for multimodal/video models can receive mixed UI signals:

• valid model interpretation output
• plus file-processing warning and/or no-citation indicator

This update documents expected behavior and reduces troubleshooting confusion.

[pr-validator-bot]

ℹ️ Documentation PR Guidelines

👋 Welcome! This is an automated message posted on all new documentation PRs to help guide our contributors. Just because this comment appeared doesn't mean you have done anything wrong!

Please ensure you're using the correct branches:

Target branch (where you're merging TO):

• dev branch: For documentation related to upcoming Open WebUI releases (new features, new environment variables, anything dependent on unreleased versions and unreleased features/fixes/changes)
• main branch: For content that can go live immediately (fixes, tutorials, documentation not dependent on unreleased features)

Source branch (where you're merging FROM):

• If targeting dev, create your branch from your fork's dev branch
• If targeting main, create your branch from your fork's main branch
• ⚠️ Mismatched branches can and will result in unwanted file changes being included in your PR!

If your docs PR depends on a pending PR in open-webui/open-webui:

• Convert this PR to DRAFT mode!
• Link to the related main repo PR in your description for clarity
• We'll review both together once the PR on the main repo is merged

Please adjust your PR target branch, source branch, and/or draft status accordingly if needed.