If it's using bipolar (or even if not) I'd want a little positive feedback to give it some hysteresis and "snap" it between on and off, so the final transistor didn't spend a lot of time unsaturated. (Think "Schmidt Trigger".) Power dissipated in the output transistor is E*I where I is current and E is voltage drop. So it dissipates a LOT more power when it's running the lamp at, say, 2 amps with 6 volts across the transistor, than when it's running with 4 amps and a half volt.
I can see how to do that with three resistors, two NPN transistors (say 2n2222 for the input stage) and the cadmium cell. (Both emitters grounded, both bases with resistor to +12, small-transistor base also has cadmium cell to ground, resistor to big transistor's collector. Big transistor base also hooked to small transistor collector. Resistive load (lamp) from big transistor collector to +12.
I think you can get away with a resistor less using a PNP / NPN pair,wired up vuagely like a split-up Triac.
And I'm SURE there's several ways to do it with an NE555, among the thousands of cute low-component-count applications for that ubiquitous el-cheapo chip. (It has a voltage divider, schmidt trigger, reference generator, and a moderately decent output transistor, all in a six-leg bug for pennies.)