Improve accuracy of summary statistic methods for transformed distributions

[mjskay]

I'm not sure the exact problem, but it seems that the mean() of transformed distributions is not always correct. Here's an example of exp() applied to dist_normal(), which should yield a lognormal distribution, giving incorrect means:

> x = dist_normal(0:1, 1:2)
> mean(x)
[1] 0 1
> mean(exp(x))   # should be roughly exp(c(0.5, 3))
[1] 1.500000 8.154845
> exp(c(0.5, 3))
[1]  1.648721 20.085537

I'm not sure the issue. It does appear to work in some cases:

> mean(log(exp(x)))    # should be roughly c(0, 1)
[1] -4.732208e-07  1.000000e+00