Pixyz

Updates
* Add the multinomial distribution (64, 67, 68)
* Add doctest (67, 69)
* Add the "product of normal distributions" classes, ProductOfNormal and ElementWiseProductOfNormal.(70)
* Add flow modules (71)
* Change the way of writing formulas to latex codes (75)
* Add a print_latex method which can display formulas of Distribution/Loss/Model API in latex format (75).

Updates
* Add Maximum Mean Discrepancy (`MMD`). (60)
* Add the Wasserstein distance (`WassersteinDistance`). (60)
* Add inverse multi-quadratic radial basis function (RBF) kernel. (60)
* Remove the `log_likelihood` method. (62)
* Remove the `NLL` class. (62)
* Add `LogProb` and `Prob` instead of `NLL`. (62)
* Add `log_prob` and `prob` methods in Distribution API each of which performs as the instance of `LogProb` and `Prob` respectively. (62)
* Rename the `LossExpectation` class to `Expectation`. (62)
* Add the `expectation` method in Loss API as the instance of the `Expectation` class. (62)
* Rename the `estimation` method in Loss API to `eval` (followed by model.eval() on PyTorch). (62)
* Remove flow implementations for now. (62)