pmReadoutSetVariance does not correctly handle low-signal high-gain

[eugene]

in pmReadoutSetVariance, there is a bit of code to keep the variance within sensible ranges: it forces the variance to be positive (taking the absolute value) and it clips the lower end to 1. the first is an acceptable hack, though it would be worth understanding how the code can ever result in a negative variance (add an assert?). the second is a bigger problem since the calculation is in DN: the clipping needs to be to the appropriate amount in DN based on the gain equivalent to a 1 e-.