Ensure computation of rhodz before dissip also when not in LAM mode

Before calling dissipation, on must compute f_mass which is used as input. Before !78 (merged) this was done in timeloop. After !78 (merged) this is done by update_border, but only when running in LAM mode.

This seems to be a bug. This is confirmed by the fact that reintroducing the computation of f_mass in timeloop changes the simulation results (trunk fails the validation pipeline). This would not be the case if f_mass were properly diagnosed elsewhere before calling dissip.