So every time I said photovoltaic I meant photoresistor.
I thought I had a LM117 laying around but I didn't so building a test in kinda hard, but by looking at the
data sheet and knowing that more light means less resistance (I got 20 ohms w/ a maglight and ~3 mega ohms under a blanket) I can kinda guess.
If photoresistor went on R1 and a variable resistor went on R2 to control the overall scale you would end up with an exponentially increasing brightness.
I would imagine that putting another variable resistor in parallel with the 2 would offer some control as to the maximum brightness and would make the whole thing have a more square root style curve (brightness increases more between dark and some ambient then it will from mid day to when light hits the sensor).
I would also imagine that putting a variable resistor with in parallel with the photoresistor would allow for an adjustable limit and a more square root style curve of the overall resistance of the photoresistor and variable resistor.
I would definitely put a zener diode between the regulated out and ground so that if the Vout is greater then 3.5 it trips and shorts out the system, it will save the LEDs but it might kill the voltage regulator. A better solution might be to run the whole system after a 3.6v regulator or something (if my memory is correct then a voltage regulator needs higher input then the output, but I might be wrong) that way the system can't fry your LEDs.
EDIT: There is only one place to put the photoresistor to get the brightness to increase with the amount of light, I removed the way that makes brightness increase in a dim room.