Here's the difference, once the plane's wings are level with the horizon, the roll is considered to have been completed. The rest (regaining a stable pitch) is recovery.
Yes, you are correct that the aircraft's velocity is not maintainable after the maneuver has been completed, and must incur positive G forces to regain level flight, but it's not technically part of the barrel roll.
EDIT: As I noted in another response (in which I go into a lot more detail), the pilot probably doesn't even have to take any action to negate the downward velocity component; the change in the angle of attack (the angle at which the wing intersects with the airflow) would naturally increase the amount of lift being generated by the wing, at the cost of more drag.
I suppose it's high-school-physics possible to fly a 1 gee helical path (the "barrel roll") centered around an orbital zero-gee trajectory...
Imagine for a moment a zero gee orbital trajectory at a low enough altitude that you can still generate aerodynamic lift from the atmosphere. (Here we handwave away all the pesky frictional heating, because we're deep in the thought experiment world of perfectly spherical cows of uniform density.) Now imagine a helical "coil spring" shaped path with that orbital trajectory running through the center. "All" you need to do is get the diameter and spacing of those helical coils right so your acceleration around the coils needs to be 1G, while your averaged out path coincides with the orbital zero gee trajectory.
<grin>
(An aircraft with sufficient speed, fuel capacity, heat shielding, and whatever else I've glossed over - is left as an exercise for the reader...)
Imagine a fighter jet flying circles around an airliner as it follows it along - with the fighter pilot flying at just the right radius and speed that it's accelerating at 1 gee for the turn (so they'd be "feeling" 2 gee as they pass under the airliner, and zero gee as they loop over the top of the airliner) using a helical "barrel roll" path - and at the same time "keeping up" with the airliner along it's path, so if you were sitting in the airliner looking out it'd look like the fighter was flying circles around the long axis of the fuselage.
Now imagine the fighter pilot does the same trick following the vomit comet - as it flies its parabolic arcs which gives it's occupants 20-30 secs or so of "zero gee".
You're neglecting two potential sources of upward acceleration. One, the turn itself, or in other words air resistance: if you stop turning when you're pointed straight up, clearly you're going to go up, not down (at least to start with), which means the turn accelerated you upward. And two, any forward acceleration provided by the engine while "forward" isn't horizontal.
(I don't know enough about aerodynamics to actually determine how a barrel roll actually works, though, only enough to contradict your post :)
You can stay arbitrarily close to 1 gee, given unlimited time and altitude, but you can't stay exactly at one gee throughout a barrel roll.