fix: add Gemma 4 model support and stop silent model ID fallback

[roomote-v0]

Related GitHub Issue

Closes: #12166

Description

This PR attempts to address Issue #12166. Feedback and guidance are welcome.

Two changes:

1. Add Gemma 4 models to the geminiModels map (packages/types/src/providers/gemini.ts):

   • gemma-4-31b-it (131K context, image support)
   • gemma-4-12b-it (131K context, image support)
   • gemma-4-6b-it (32K context, image support)
   • gemma-4-4b-it (32K context, image support)
   • gemma-4-1b-it (32K context, text only)

2. Fix silent model ID fallback in both GeminiHandler.getModel() and VertexHandler.getModel():

   • Previously, any model ID not in the known models map was silently replaced with the default (gemini-3.1-pro-preview), causing 429 quota errors when users specified custom models like Gemma 4
   • Now the user-provided model ID is always preserved; only the model info (context window, pricing, etc.) falls back to defaults for unknown models
   • This follows the same pattern used by the DeepSeek provider

Test Procedure

• Ran cd src && npx vitest run api/providers/__tests__/gemini.spec.ts api/providers/__tests__/vertex.spec.ts -- all 28 tests pass
• Updated existing test that expected unknown model IDs to be replaced with default
• Added new tests verifying:
  • Unknown model IDs are preserved (not silently replaced)
  • Default model is used when no model ID is provided
  • Gemma 4 models are natively recognized with correct context window and capabilities

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes.
• Documentation Impact: No documentation updates are required.
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Documentation Updates

• No documentation updates are required.

Additional Notes

The fix applies the same pattern used by the DeepSeek provider (deepseek.ts:37-48), which preserves the user-provided model ID and only falls back on model info for unknown models.

Interactively review PR in Roo Code Cloud