feat(qwen): migrate Qwen provider to AI SDK

[daniel-lxs]

Summary

Migrates the Qwen Code provider to use the Vercel AI SDK (@ai-sdk/openai-compatible) instead of the direct OpenAI SDK.

Changes

src/api/providers/qwen-code.ts

• Replaced direct OpenAI SDK usage with @ai-sdk/openai-compatible's createOpenAICompatible
• Now uses streamText and generateText from the ai package for streaming and completion
• Uses shared transform utilities (convertToAiSdkMessages, convertToolsForAiSdk, processAiSdkStreamPart, mapToolChoice)
• Maintains OAuth credential management with automatic token refresh
• Provider is lazily created and recreated when credentials change
• Added getModelParams integration for proper model parameter handling

src/api/providers/__tests__/qwen-code-native-tools.spec.ts

• Updated mocks to use AI SDK mocking patterns (mockStreamText, mockGenerateText)
• Added mock for @ai-sdk/openai-compatible's createOpenAICompatible
• Updated test events to use AI SDK stream event types (tool-input-start, tool-input-delta, tool-input-end, reasoning)

Benefits

• Consistent with other AI SDK-based providers (DeepSeek, Moonshot, HuggingFace)
• Uses shared transform utilities for message and tool conversion
• Proper handling of tool calls via AI SDK's tool streaming events
• Reasoning content handled via AI SDK's reasoning events
• Maintains backward compatibility with existing OAuth credential flow

Testing

• All 13 tests pass
• Lint passes
• TypeScript compilation passes

Resolves EXT-712

---

Important

Migrates Qwen Code provider to Vercel AI SDK, updating message handling, tool integration, and OAuth management.

• Behavior:
  • Migrates Qwen Code provider in qwen-code.ts to use Vercel AI SDK (@ai-sdk/openai-compatible) instead of OpenAI SDK.
  • Uses streamText and generateText for streaming and completion.
  • Integrates getModelParams for model parameter handling.
  • Maintains OAuth credential management with automatic token refresh.
  • Provider is lazily created and recreated when credentials change.
• Tests:
  • Updates qwen-code-native-tools.spec.ts to use AI SDK mocking patterns (mockStreamText, mockGenerateText).
  • Adds mock for createOpenAICompatible.
  • Updates test events to use AI SDK stream event types (tool-input-start, tool-input-delta, tool-input-end, reasoning).
• Misc:
  • Uses shared transform utilities for message and tool conversion.
  • Handles tool calls via AI SDK's tool streaming events.
  • Handles reasoning content via AI SDK's reasoning events.

^{This description was created by for f9c432e. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Rooviewer   See task on Roo Cloud

Review complete. Found 1 issue to address:

• Add error?.status === 401 to the 401 error detection check in createMessage and completePrompt methods (AI SDK uses .status as the primary property for AI_APICallError)

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[roomote]

The 401 error detection may not work correctly with AI SDK errors. AI_APICallError uses .status as its primary property, and AI_RetryError exposes it via .lastError?.status. The current check only looks at .statusCode and .response?.status, which means token refresh may not trigger when the API returns 401.

Suggested change

	if (error?.statusCode === 401 || error?.response?.status === 401) {
	if (error?.status === 401 || error?.statusCode === 401 || error?.response?.status === 401) {

The same pattern applies to the completePrompt method at line 347.

_{Fix it with Roo Code or mention @roomote and request a fix.}

[daniel-lxs]

This is Qwen Code not Qwen itself