Pyrichlet

This is a minor release to improve the `gibbs_map_pairplot` function by adding tick labels to diagonal plots, it also updates and improves the documentation.

Changes

- Improve `gibbs_map_pairplot` diagonal plots by adding the mixture density as well as unused components to the plot.
- Tests: Change parameters for broader testing.

Bug fixes

- Fix function `mixture_density` not scaling components by their respective weighting.
- Fix `BetaInDirichlet` weighting model's method `random` not extending the array `_v_base` of base sticks.

Bug fixes

- Fix bug in data loaders `load_chaetocnema` and `load_penguins`.

Bug fixes
- Fix missing package data due to an incomplete MANIFEST.in file which prevented the package from loading in the previous version.

Improvements

- Data loaders `load_chaetocnema` and `load_penguins` were added in `pyrichlet.utils`
- Added arguments to density estimators to get collapsed densities or component densities
- Refactor weighting model rng state as an internal variable
- Change the parameters for `fit_variational` to be more consistent with `fit_gibbs`
- For `BetaInDirichlet`, implement a marginal sampler for weight fitting.

Bug fixes

- `BetaBernoulli` weights were wrongly indexed to its persist vector element.

Improvements

- Refactor code to follow best practices
- Check if mixture class object is fitted (Gibbs, Variational)
- k-means initialization now returns assignations ordered by size

Bug fixes

- Correct error in posterior variance calculation
- Fix fitting bugs in the (Geometric, Frequency and Equal)Weighting
- Fix theta update calculation in variational fitting