Simplify...
Conveyor between two planets (or just between to locations, does not matter whether planets or just locations - all that is specified is a distance agreed on by the army and the conveyor at rest)
Army of clock wearing roller skaters (does not have to be roller skates, probably better to just assume the army is floating in formation arbitrarily close to the conveyor, the conveyor has a length and the army has a length, both the same)
The conveyor begins to move, the army stays in place...
The conveyor movement approaches c... this is the first tricky part concerning Frames of reference:
FOR Army
FOR Upper belt of conveyor
FOR Lower belt of conveyor
FOR Planet Left
FOR Planet Right
Army folks see the upper and lower belts become length contracted. The question about "levitation" is just an awkward word - it means will the army continue to fit on the the upper belt or be crowed off of it? Army folks will see the lower belt contract same amount as the upper belt. Army will observe that the belt no longer extends all the way to either planet.
Upper and lower belts will see the army contract... and each belt will see the other belt contract even more than the army contraction, and each belt will see the planets contracting.
Each of the planets will see both upper and lower belts contracting.
The question is about how to reconcile the various length distortions. When an observer at rest is observing a fast mover, the usual way to "measure" is to use the transforms to convert the observed distortions to local values for the resting observer, but in this case it is as if the resting observer has an indefinitely long measuring rod that extends to the place of the fast mover - and the rest observer is measuring by this long rod as an extension of his FOR... I think this is the problem with the problem.
This problem is similar to the "Lightspeed Submarine Paradox" where a submerged sub is approaching lightspeed. The FOR from the sub sees the water rushing past gaining mass and conclude the sub should become more buoyant and surface. Observers on the coast see the sub as gaining mass and should dive. Something has to happen...?
But maybe "something happening" is always local? Maybe the resolution to both is that relativistic effects require distance... measurements of events (distant FOR) made from a distance (local FOR). Local measurements will always be non-relativistic and relativistic measures from a distance will have no bearing on the local mechanics of what is happening?
In which case, the army of skaters will have plenty of room on the conveyor, and I guess the sub will surface...?