should add hard limits in connection with distribution functions

[danielskatz]

Right now, we have a lower limit of zero set for distributions. For example, if I create a skeleton with input file size that's a Guassian distribution around 1 MB with stddev of 0.5 MB, some values will be negative, which doesn't make sense, so we set those to zero.

What we really should do is optionally accept hard limits along with distribution mean and stddev, and when values are generated outside of those limits, the values would then be set to the limits.