is it counted as independent worker request with subrequest and cpu costs or is it charged only as part of compute on the worker calling it ?
I'm asking this because if we use it as interpreter for LLM generated code it'll be huge cost if it costs a seperate worker request per LLM call