Add general non-parametric bootstrapping

[ianfiske]

This should work for any unmarked fitting function. A general design idea is a master nonparboot function that requires each fitting function provide:

1. A resample function to properly resample the data
2. A fitting function that properly assembles calls to the fitting function
3. An extractor function that pulls out the desired stats

The master nonparboot function should then do the right thing with these bootstrap samples.

nonparboot should take a fit object and return the fit object with standard error and associated stats modified and also additional bootstrap stats (maybe just num. iterations and raw b.s. stats?) tacked on. We should create a new class for these fit objects and the results should inherit from both these and the original fit class.

[ianfiske]

Improved docs for nonparboot and added examples. Closed by 80e669758316b1948a4365dfdf4d5ed465d1d7e8.