fix(core): fail closed when Gemini RPD counter is unavailable (#72)

[cursor]

What

• Make bot and quiz Gemini RPD counter increments fail closed when DynamoDB cannot update the counter.
• Map counter-unavailable failures to existing provider-unavailable errors so callers use the established fallback/error paths instead of calling Gemini unmetered.
• Add regression tests proving DynamoDB update failures do not proceed to Gemini API calls.

Why

• A DynamoDB UpdateItem failure previously returned (0, True), bypassing the daily Gemini RPD guard.
• Concrete impact: table outage, IAM misconfiguration, throttling, or region/account mismatch allowed unbounded Gemini requests with no counter movement, defeating quota controls and fallback policy.

Verification

• python3 -m pytest tests/test_rate_limit_fail_closed.py tests/test_explain_multimodal.py tests/test_quiz_generator.py
• python3 -m pytest

📝 Description

Fail closed when Gemini RPD counters cannot be incremented safely.

🚀 Type of Change

• 🐛 Bug Fix (Fix an issue)
• ✨ New Feature (Add a new feature)
• 🏗️ Infrastructure (CDK/AWS resource changes)
• ♻️ Refactoring (Code refactoring, no logic changes)
• 📚 Documentation (Documentation update)

🏗️ Infrastructure Changes (Crucial)

• Resources Created: None
• Resources Modified: None
• Resources Deleted: None
• Cost Impact: Prevents unexpected Gemini usage when DynamoDB quota accounting is unavailable.

✅ Checklist

• I have performed a self-review of my code.
• I have run pre-commit run --all-files locally and fixed all issues.
• I have added/updated Unit Tests (pytest).
• My changes do not generate new warnings.
• (If Infra change) I have run cdk diff locally to verify changes.

🔗 Related Issues

• Fixes feat(bot): explain multimodal + document auto summary #72