execute_command tool should report (to the assistant/model) command run time

[TaoK]

On certain kinds of tasks, the commands the assistant runs can easily end up running long, and needing to be "killed" by the user. When this happens, the assistant typically just sees an Exit code, and no output. It then often/typically "flails", immediately trying to change some arbitrary thing in the absence of any information about what went wrong, and going off in a bad direction. The user then needs to explicitly cancel and send a message explaining what happened / what they did (if they're paying attention, which they ideally shouldn't need to be!)

This can be improved to some extent with explicit prompting (eg "If a command fails without output, consider that I may have killed the process because it was running too long"), but if commands were to routinely output their run time / duration, this would be both a strong implicit hint (a model can "reason" about an arbitrary failure that occurred exactly at the 180-second mark as being a likely timeout, for example), and would also allow for better explicit user-defined prompts/rules eg in AGENTS.md, eg saying "If a command ran longer than 10 seconds and failed without any error message output, consider that I may have killed the process; ask if in doubt."