[BUG] Gemini: Local cost calculation ignores tiered pricing and cache discounts

[ikumi3]

Problem (one or two sentences)

When using Gemini through Vertex AI, tiered pricing and cache discounts are not applied in the local cost calculation, causing displayed costs to always show the highest tier rates instead of the correct discounted rates.

Context (who is affected and when)

Who is affected:

• All users using Gemini models through Vertex AI
• Users monitoring costs through Roo-Code's local logs and UI
• Users expecting accurate cost tracking and reporting

When this occurs:

• When using Vertex AI provider with Gemini models
• During local cost calculation in the calculateCost method
• Affects cost display in ui_messages.json and local cost tracking
• Note: Actual billing by Google Cloud/Vertex AI remains correct

Reproduction steps

1. Configure Roo-Code to use Vertex AI with a Gemini model (gemini-2.5-pro)
2. Set up Vertex AI credentials (project ID, region, authentication)
3. Make an API call that generates significant token usage
4. Check the cost displayed in ui_messages.json or local logs
5. Compare with expected cost calculation using tiered pricing and cache discounts

Expected result

Displayed cost should reflect tiered pricing and cache discounts

Actual result

Displayed cost always shows high-tier pricing applied to all tokens

Variations tried (optional)

No response

App Version

v3.28.1

API Provider (optional)

Google Vertex AI

Model Used (optional)

gemini-2.5-pro

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I'm exploring a potential fix for the Gemini tiered pricing and cache discount calculation issue. I'll investigate the cost calculation logic for Vertex AI and work on implementing proper tiered pricing support.

[roomote]

I've opened PR #8018 with a fix for the tiered pricing calculation issue: #8018

The fix modifies the calculateCost method in GeminiHandler to properly handle models where cache read prices are only defined within tiers. This ensures that Gemini models via Vertex AI will now correctly apply tiered pricing based on token usage.

The PR includes comprehensive tests to verify the tiered pricing works correctly for different token thresholds. Most CI checks have passed, though there's a Windows test failure that appears to be unrelated to this change.