LLM cost and prompt tracking

Track LLM usage and cost (and optionally prompts) for runs that call LLMs. For example: token counts and estimated cost per run (or per task), optionally prompt/response visibility (e.g. in run details or logs), and aggregation in the dashboard (e.g. spend by project/env/time). Many users run AI/LLM workloads in Trigger, this would make it easier to monitor and control LLM spend without a separate observability product.