Pyrichlet

Improvements

- Implement variational methods for DirichletDistribution
- Implement variational methods for DirichletProcess
- Implement variational methods for PitmanYorProcess and add truncation optional parameter
- Add FrequencyWeighting class
- Implement variational methods for FrequencyWeighting (this emulates the EM algorithm since FrequencyWeighting is not a Bayesian model)
- Add EqualWeighting class
- Add a mixture model for each weighting model

Bug fixes

- GeometricProcess weighting model only worked with a hardcoded number of groups

- License changed to Apache License 2.0
- Add `fit_variational` method to the `BaseGaussianMixture` class to aproximate the posterior distribution via variational inference.