Track AI API costs via centralized query service

[mbuckbee]

Summary

Route AI API calls through a centralized service that calculates and tracks token costs per request. Instead of each component calling AI providers directly and discarding token counts, funnel calls through a query service that:

1. Calls the underlying provider (OpenRouter, Gemini, etc.)
2. Looks up the model in a cost table
3. Calculates cost from input/output token counts and model pricing
4. Returns the response with cost_in_dollars attached

Motivation

Currently AI API calls happen directly with no cost visibility. This makes it impossible to understand per-video generation costs or identify expensive operations.

Approach

• Add a general-purpose query method to a central AI service that wraps any provider
• Track input_tokens, output_tokens, cost_in_dollars, and model_slug on the relevant record
• Accumulate costs across multiple AI calls within a single operation
• Persist totals after successful completion

Reference

Implemented in Knowatoa's AiQueryService.query + TaskExecutable concern as a working pattern.