Toward analytic theory of the Rayleigh–Taylor instability: lessons from a toy model