It occurs to me that to control the output of a current source you need a (remote-controlled) variable shunt resistor across it, with the resistor being "off" during startup.
I don't know how well the ballast will tolerate current draws above or below design values. You may need some sort of an L network or a T network that attenuates current to the tube but so the ballast sees a constant current load but presents another current source of lesser value to the tube.
You need the dual of a network that does this with voltages, in the sense that a current source is the dual of a voltage source.
Good problem!!!
I'd start by searching online patents to get some idea of the complexity you'll be looking at.