It is a common misconception that the vertical CG position (relative to the wing) plays a major role in roll stability. It doesn't; it's mostly about aerodynamics. (You mention it yourself in "as a result of the differences in localized airflow...")
There are other important factors at play as well: dihedral angle, wing sweep, design of the vertical stabiliser.
Presumably, you are interested in the behaviour of an aircraft with the perfectly neutral roll stability. If so, it needs at least these conditions:
- General symmetry about the horizontal plane, which includes:
- Zero wing dihedral;
- Symmetric (e.g. round) fuselage;
- Wing at the centre of fuselage (vertically);
- Symmetric vertical stabiliser (e.g. with dorsal fin);
- Absolute rigidity so that the symmetry remains under load.
- Zero wing sweep (straight wing);
- Lack of any artificial stability augmentation.
(Technically speaking, the geometric symmetry is not necessarily required: we can compensate one effect with another, say, wing sweep with anhedral. But in the spirit of the original question, let's assume the full symmetry).
We should also assume a reasonable stability in pure yaw.
So, if you induce a pure roll disturbance and then remove it,
- the aircraft will start rolling.
Because it has some inertia about the roll axis, it will not simply stop when you remove the disturbance. The only force that will counteract the roll now is roll damping. It naturally occurs because the descending wing has higher angle of attack (and vice versa). The strength of the effect depends on the square of the wingspan, but in theory the spherical cow will keep turning forever, albeit with ever-diminishing speed.
So far, nothing would induce a yaw, and the airflow would still be symmetric.
However, as roll develops, and the aircraft keeps the same pitch balance, the lift will become insufficient to keep it level. The usual thing happens:
- the aircraft enters a spiral dive.
Now, this is not a symmetric situation anymore. There is a side force from gravity; sideslip will develop; the aircraft will yaw due to its yaw stability that we assumed. Things start to get complicated (that is, 'normal'), with many variables involved. Even in the vertical plane, despite postulated symmetry, the shadowing and other cross-flow effects on the top and bottom of the wing will be different (the wing, by definition, cannot be aerodynamically symmetric when it generates lift), and thus some roll effect may be observed.
So it's hard to predict further behaviour in general. At one extreme, with stubby wings and slender body and very strong initial disturbance, the aircraft may barrel roll many times and may even enter a peculiar stable roll motion due to inertia coupling. At the other, the aircraft will just roll a little and will enter a gentle (nearly) coordinated turn with very slight descent.