implement loadPow

Jack Lewis writes: "i have found that fitting power models using non-linear least squares has big advantages over using log-transformed responses: (1) you can use all the zeroes in both model-fitting and prediction, (2) no bias-correction is needed, and (3) power models nearly always fit the higher concentrations more closely, since the regression line isn't prone to being levered by an abundance of small values that are widely dispersed in log space. With R it's a snap. `powfit <- nls(c ~ a*q^b, start = list(...))` Starting values come immediately from log-log regression."

What a great idea! We should do it.