fix: apply tiered pricing for Gemini models via Vertex AI

[roomote]

Summary

This PR fixes the local cost calculation for Gemini models when used through Vertex AI, ensuring that tiered pricing and cache discounts are properly applied.

Problem

When using Gemini models through Vertex AI, the local cost calculation was always showing the highest tier rates instead of applying the correct tiered pricing based on token usage. This was causing discrepancies between the displayed costs in Roo-Code and the actual billing from Google Cloud.

Solution

Modified the calculateCost method in GeminiHandler to:

• Properly handle models where cacheReadsPrice is only defined within tiers (not at the top level)
• Apply tiered pricing based on input token count
• Gracefully handle missing prices with appropriate fallbacks

Changes

• Modified src/api/providers/gemini.ts: Updated calculateCost method to handle tiered pricing correctly
• Added src/api/providers/__tests__/vertex-tiered-pricing.spec.ts: Comprehensive test suite for Vertex AI tiered pricing

Testing

• ✅ All new tests pass for tiered pricing scenarios
• ✅ Existing Gemini provider tests continue to pass
• ✅ Verified correct pricing for tokens under 200K (tier 1) and over 200K (tier 2)
• ✅ Cache read pricing correctly applied based on tier

Impact

This fix ensures that users see accurate cost estimates in their local logs and UI when using Gemini models through Vertex AI. The actual billing by Google Cloud was already correct - this only affects the local display.

Fixes #8017

---

Important

Fixes local cost calculation for Gemini models via Vertex AI by implementing tiered pricing in gemini.ts and adding comprehensive tests.

• Behavior:
  • Fixes local cost calculation for Gemini models via Vertex AI to apply tiered pricing and cache discounts.
  • Handles models with cacheReadsPrice defined only within tiers.
  • Applies tiered pricing based on input token count and handles missing prices with fallbacks.
• Code Changes:
  • Updates calculateCost in GeminiHandler in gemini.ts to handle tiered pricing.
  • Adds vertex-tiered-pricing.spec.ts for testing tiered pricing scenarios.
• Testing:
  • New tests verify tiered pricing for tokens under and over 200K, and cache read pricing.
  • Ensures models without tiered pricing use flat rates.
  • All tests pass, ensuring accurate cost estimates in local logs and UI.

^{This description was created by for d53ece6. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

I reviewed my own code and found it surprisingly coherent. Must be a bug in the review process.

Review Summary

The fix correctly implements tiered pricing for Gemini models via Vertex AI. The logic properly handles models with and without tiers, and gracefully handles missing cache prices.

Suggestions for improvement:

1. src/api/providers/gemini.ts:313 - When cacheReadsPrice is undefined, we're setting it to 0. Consider adding a comment explaining this is intentional for models without cache support.

2. src/api/providers/gemini.ts:297 - The tier selection uses find() which returns the first matching tier. Consider adding a comment clarifying that tiers must be ordered from lowest to highest context window.

3. src/api/providers/tests/vertex-tiered-pricing.spec.ts - Consider adding a test case for when cacheReadTokens is provided but cacheReadsPrice is undefined to explicitly test the fallback behavior.

What works well:

• Comprehensive test coverage that validates both tiered and flat pricing scenarios
• The test at line 35-37 effectively validates that tier 1 pricing is used for tokens under 200K, directly addressing the reported bug
• Proper handling of edge cases where prices might only be defined in tiers

Overall, this is a solid fix that addresses the issue reported in #8017.