How to stop/cancel a cuda kernel launched by Numba before it ends?

Question

I have a simulation written with Python/Numba that uses several cuda GPUs. Each one is launched from a different process using a separate cuda context. This simulation runs