Bipartitepandas

- Update clustering so you can specify the outcome column to use for clustering data

- Vectorize Quantiles code

- Add new connectedness option strongly leave out x, which computes the largest set of observations that is strongly connected, and also leave-one-x-out connected (so if an observation/spell/match/etc. is dropped, it remains connected, but not necessarily strongly connected). This constructs a dataset that can run the Sorkin revealed preference estimator as well as the KSS HE bias correction.
- Clean up and vectorize Measures code (and fix some bugs with quantile_firm CDF option)

-Add classes `BipartiteExtendedEventStudy` and `BipartiteExtendedEventStudyCollapsed`

-Remove method `.get_extended_eventstudy()` from `BipartiteLong` and replace with `.to_extendedeventstudy()` in `BipartiteLongBase`

-Remove method `.plot_extended_eventstudy()`, will move to PyTwoWay

-Move class ParamsDict into a separate package (available [here](https://github.com/adamoppenheimer/paramsdict))

-Delete graphs created while computing largest leave-one-out set (hopefully this saves some memory)