Expose thinking tokens for roo/sonic

[cte]

Important

Enhance RooHandler to expose reasoning tokens, update model description, and add dynamic provider URL configuration.

• Behavior:
  • RooHandler in roo.ts now yields reasoning tokens in createMessage().
  • Updates rooModels in roo.ts to support prompt caching and modifies the model description.
  • Adds ROO_CODE_PROVIDER_URL to .env.sample for dynamic base URL configuration.
• Refactoring:
  • Extracts createStream() from createMessage() in base-openai-compatible-provider.ts for better code organization.
• Tests:
  • Updates roo.spec.ts to test new reasoning token behavior and prompt cache support.

^{This description was created by for 350bab4. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Thank you for your contribution! I've reviewed the changes and found some areas that need attention.

Critical Issue:

• The test file src/api/providers/__tests__/roo.spec.ts needs to be updated to include test coverage for the new reasoning_content functionality. This is a significant feature that requires proper testing.

Other feedback provided inline on the changed files.

[mrubens]

@roomote-agent please fix the failing test

[roomote]

Hi @mrubens! I see the request to fix the failing test. Let me investigate the issue and fix it for you. 🔧

[roomote]

Hi @mrubens! ✅ I have successfully fixed the failing test.

Summary of the fix:

The test "should handle unknown model ID with fallback info" in src/api/providers/tests/roo.spec.ts was failing because the RooHandler.getModel() method was incorrectly returning the roo/sonic model info as fallback for unknown models.

What I changed:

• Updated src/api/providers/roo.ts line 74 to remove the fallback to rooDefaultModelId model info
• This ensures unknown models get proper fallback info with supportsPromptCache: false (matching test expectations)

Test results:

✅ platform-unit-test (ubuntu-latest) - Now passing (was failing before)
✅ All provider tests passing locally

The integration test failure appears to be unrelated to this change and was already failing before my fix.

Commit: 7ed6620