Fixes #92688.

This updates the media-understanding image runtime so Qwen/DashScope Model Studio image requests put the prompt text in the user message with the image block. DashScope rejects the current system-prompt plus image-only user message shape with Unexpected item type in content.

The change should stay narrow: extend the prompt-in-user-content routing for modelstudio-native image media-understanding endpoints and add a focused regression asserting Qwen/DashScope image requests have no systemPrompt and include { type: "text" } before the image content.

Credit: thanks @Yachiyo404 for the report and reproduction details in #92688.

Clownfish 🐠 replacement reef notes:

• Cluster: gitcrawl-125-autonomous-terminal-gap
• Source PRs: none
• Credit: Credit @Yachiyo404 for the [Bug]: Qwen vision models fail with 400 "Unexpected item type in content" on DashScope #92688 report and reproduction details.; Do not treat Qwen: qwen3.6-plus image understanding blocked on Coding Plan endpoint despite model supporting vision #63654 or WebChat image recognition fails when using coding.dashscope base URL for Qwen models #67759 as duplicates to close; both are already closed older model-resolution issues.; Existing-overlap PR refs #92688: fix(qwen): use DashScope native image format for Qwen vision models #92704, fix(media-understanding): place Qwen/DashScope image prompts in user content (#92688) #92770, and fix #92688: [Bug]: Qwen vision models fail with 400 "Unexpected item type in content" on DashScope #92782 were excluded from actionable refs in this job and are not source PRs for this replacement artifact.
• Validation: pnpm -s vitest run src/media-understanding/image.test.ts src/agents/tools/image-tool.test.ts; pnpm check:changed

fish notes: model gpt-5.5, reasoning medium; reviewed against f114920.