I remember that we fixed the issue with the missing 0.5*h*F some years ago. For a flat bounce-back wall aligned with the lattice axis, the actual hydrodynamic no-slip wall is positioned in the middle between the last fluid node and the first wall node. The (0, 0, 0) node of the LB lattice in Espresso is placed at (agrid/2, agrid/2, agrid/2). With that information you should be able to figure out the "real" width of your channel. Does that make it better?
The comment on being able to determine the position of the wall down to 1 agrid only applies to walls not aligned with the lattice axis, like the ones I used in the B.Sc. thesis.