Text this: A reinforcement learning strategy to automate and accelerate h/p-multigrid solvers