fix: make embedding model ID matching case-insensitive

[roomote]

This PR addresses Issue #9026 where codebase indexing fails with custom providers using different model ID casing.

Problem

Users with custom providers (e.g., OpenRouter) using model IDs with different casing than defined in our profiles (like Qwen/Qwen3-Embedding-8B vs qwen/qwen3-embedding-8b) were experiencing indexing failures. While regular Roo Code operations worked, the embedding model dimension lookup failed silently, preventing codebase indexing.

Solution

• Updated getModelDimension(), getModelScoreThreshold(), and getModelQueryPrefix() in embeddingModels.ts to perform case-insensitive model ID matching
• Uses Object.keys().find() with .toLowerCase() for comparison while preserving original profile keys

Testing

• Added comprehensive test suite in embeddingModels.spec.ts with 22 tests
• Tests cover all three functions with various case permutations
• Includes specific tests for the exact Qwen model casing reported in the issue
• All existing tests continue to pass, ensuring backward compatibility

Impact

• Fixes the reported indexing issue for custom providers using different model ID casing
• Improves user experience by making model configuration more forgiving
• Maintains full backward compatibility with existing lowercase model IDs
• No security implications or breaking changes

Fixes #9026

Feedback and testing welcome! This should resolve the indexing issue, but if the problem persists, there may be additional factors to investigate.

---

Important

This PR makes model ID matching case-insensitive in embeddingModels.ts to fix indexing issues with custom providers, adding comprehensive tests for validation.

• Behavior:
  • getModelDimension(), getModelScoreThreshold(), and getModelQueryPrefix() in embeddingModels.ts now perform case-insensitive model ID matching using .toLowerCase().
  • Uses Object.keys().find() to preserve original profile keys.
• Testing:
  • Added 22 tests in embeddingModels.spec.ts to cover case-insensitive matching for all three functions.
  • Includes tests for specific Qwen model casing and various case permutations.
• Impact:
  • Fixes indexing issue for custom providers with different model ID casing.
  • Maintains backward compatibility with existing lowercase model IDs.

^{This description was created by for 96dabb5. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Review complete. The implementation correctly addresses the case-insensitive model ID matching issue with comprehensive test coverage. No issues found.

_{Mention @roomote in a comment to trigger your PR Fixer agent and make changes to this pull request.}

[ellipsis-dev]

Consider renaming the variable used for the found key in getModelDimension from modelProfile to something like modelProfileKey for consistency with the similar functions below.

[ellipsis-dev]

The case-insensitive key lookup logic is repeated in getModelDimension, getModelScoreThreshold, and getModelQueryPrefix. Consider extracting this common logic into a helper function to reduce duplication.