Xlogit

- Tolerance parameter added to MultinomialLogit fit
- Fix bug with pre-processing of avail input

- Added batch processing to save GPU memory. The new version can estimate Mixed Logit models using hundreds of thousands of random draws in minutes
- Added parameter `halton_opts` to `MixedLogit.fit` method to allow setting the Halton draws options
- Added parameter `top_opts` to `MixedLogit.fit` to set the tolerance of for estimation
- Default number of Halton draws changed to 1,000
- Updated benchmark results

- Fix bug in `predict` when using GPU
- Fix bug in `wide_to_long` when `varying` is None.

- Predict function added
- Utility to convert from wide to long data added
- Utility for likelihood ratio test added

- Support for availability of alternatives for choice situations added
- Documentation and examples expanded

- Additional mixing distributions implemented
- Handling of unbalanced panels
- First version with read-the-docs documentation