Per-round token use for one run

The run's stored per-round model usage, in round order: model, uncached input, cache read, cache write and output tokens, an estimated cost in micro US dollars at current prices, which is not the recorded charge, and the prompt record (call kind, whether it was the final call, and the names and sizes of the system blocks). The estimate is empty for a model the price list does not know. rounds_used counts metered agent rounds when available; a null value means the count is unknown. complete is false when the history kept fewer rounds than rounds_used, because it keeps the first 1,024. Auxiliary model calls are not rounds and are not listed. No prompt or message text is returned.