fix: replace unsupported 'file' content type with text fallback for PDF (#1832)

[echoVic]

Problem

When read_file reads a PDF file, the OpenAI content converter generates content parts with type: 'file', but DashScope/Qwen API only supports text, image_url, video_url, and video content types. This causes a 400 error:

Error: Internal error (code: -32603)
Data: {"details":"400 Invalid value: file. Supported values are: 'text','image_url','video_url' and 'video'."}

Root Cause

In converter.ts, the createMediaContentPart method converts PDF inlineData and fileData parts to type: 'file' format (OpenAIContentPartFile), which is not supported by the DashScope API.

Fix

• Replace PDF content parts with a descriptive text fallback instead of the unsupported type: 'file' format
• Remove the unused OpenAIContentPartFile type definition
• Update converter tests to match the new behavior

All 49 existing tests pass.

Fixes #1832

[tanzhenxin]

@echoVic Thank for your contribution! But we're going with #1982 because it solves the same problem more comprehensively: instead of blocking PDFs entirely, it assigns appropriate modalities per model. This means we can support it once a model can read PDFs natively.

Thanks again for taking the time to dig into this and write such a thorough PR!