Add "audit mode" to manually approve tool calls before they execute

[g-linville]

It would be great if there was a command line argument that would cause gptscript to execute in an interactive mode, where the user is prompted to manually approve or reject each tool call before it is executed. This would make me more confident when testing gptscripts with access to very permissive tools (like sys.exec).

[g-linville]

I think we have had this feature for a while now but decided to call it "confirm" instead. Closing.