AI & LLM usage

Model spend, tokens, and latency on master over the last 6h — broken down by model, org, and operation. Aggregated live from the BigQuery usage logs (org-level; the usage log carries no per-user id).