You're right. The symmetry control selects Delta Vc, and this adds to one source and subtracts from the other, to keep the overall frequency nearly constant.
Incidentally, I think there may be a sign mistake on the formulas at U2A and U2B non-inverting input. I would expect it to be (Vref -DeltaVc)/2 and (-Vref-Vc)/2 respectively. Unless I'm missing something.
Similarly on the inverting inputs, I would expect to see (Vref -IsrcRsense-Vc)/2 and (-Vref+IsinkRSense+VC)/2.
Nope. Remember that a voltage divider between two potentials averages them - a sum and then a division.
U2A inverting: voltage divider between Vc and (Vref - IsrcRsense), giving the average of the two: (Vc + Vref - IsrcRsense)/2
U2A noninvering: v. div. between DVc and +Vref, giving (DVc + Vref) / 2
U2B noninverting: v. div. between Vc and -Vref, giving (Vc + -Vref)/2 = (Vc - Vref)/2
U2B inverting: v. div. between -DVc and IsinkRsense-Vref, giving (IsinkRsense-Vref-DVc)/2
The op amps won't saturate. Consider U2A. Assume Vc = 3V, DVc = 0V, and there's only 500uA flowing (too little: 3V/3k9 should give 769uA). Then the voltage at the inverting input will be higher (lower drop in R19, bringing it nearer to Vref). This will cause the output to drop, increasing the (negative) base-emitter voltage of Q1 and increasing the current.