Consistent sampling features across methods

[jdherman]

There are many nice sampling features, but they are not always implemented consistently across methods.

• Group sampling:
  Currently works for Sobol and Morris. Is it possible to implement for other methods?

• Nonuniform sampling:
  Currently works for Sobol, with several others in progress Feature - non uniform sampling - DNM #192
  The distributions are uniform, triangular, normal, and lognormal. Consider others? Maybe the option to pass any CDF function from scipy into the sampler. That could remove the confusing bounds parameter, where the parameters for the other distributions are not really bounds.

Building on the ideas in #234 , it might be possible to refactor these sampling features to avoid duplicate code and improve compatibility across methods.