- Added fallback logic to get provider/model combinations from token_usage table
- Added context dimension recording in request success path
- Model performance now displays data even if context_dimensions table is empty
- Skip providers with zero requests in selected time range