Ramsey policy: compute FOC *after* removing leads/lags of two or more.

Otherwise, on big models with lots of leads/lags (like GPM7), the Lagrangian
can be so huge that the preprocessor requires an insane amount of memory.