Theoretical time evolution of numerical errors when using floating point numbers in shallow‐water models