Samurai007AK/gpumesh ? reverse-engineered prompt

Build me a Python tool called gpumesh that lets me borrow GPU power from other machines on my network and use them like one shared pool. I want a simple command line setup flow where one machine becomes the coordinator and other machines can join as workers, with auto discovery on the same network, a shared token for access, and a friendly interactive wizard for setup. It should handle worker dropouts gracefully and re queue tasks if a machine disappears.

I also want a clean Python API so I can connect to a coordinator and decorate a normal function with something like accelerate, then call it once to run on the best local device or call map with many payloads to spread work across all connected devices. If the mesh is unavailable, it should quietly fall back to local execution. Please include commands for starting the coordinator, joining workers, seeing connected workers and devices, submitting jobs, checking status, canceling jobs, and showing saved connection info. Look up current docs online if you need to.