Return token usage per function call

[kamath]

We need better telemetry on how many input/output tokens went into each Stagehand function so we can help people better measure cost with Stagehand and help facilitate cost-benefit analysis of using smaller/cheaper models over flagship models

[ConProgramming]

Definite plus one here. It'd be fantastic to be able to do a .totalCost() at the end of a run, to see how economical using Stagehand for scale is.