Option mfs=3 gives wrong result

The mfs=3 option of the model keyword gives wrong results.

This can for example be verified by adding mfs=3 to deterministic/lola_solve_one_boundary.mod.

Once this is fixed, we should add tests that check that all the possible values of mfs give the same result on a given model.

The option mfs is also broken in version 4.6.1. Several other problems with mfs have already be fixed in the 4.6 branch, but the present issue is still unfixed.