Irspack

What's Changed
* Version 0.2.0 by tohtsky in https://github.com/tohtsky/irspack/pull/89
* Drop Python3.6 support & Fix test by tohtsky in https://github.com/tohtsky/irspack/pull/90

**Full Changelog**: https://github.com/tohtsky/irspack/compare/v0.1.20...v0.2.0

Breaking Changes
- iALS's confidence parameter `alpha` is now reparametrized by `alpha0 = 1 / alpha` (roughtly) following ["Revisiting the Performance of iALS on Item Recommendation Benchmarks"](https://arxiv.org/abs/2110.14037)
- Dropped the user cold start module.

What's Changed
* Improve wheel build by tohtsky in https://github.com/tohtsky/irspack/pull/88

**Full Changelog**: https://github.com/tohtsky/irspack/compare/v0.1.19...v0.1.20

What's Changed
* Improve id mapping by tohtsky in https://github.com/tohtsky/irspack/pull/87


**Full Changelog**: https://github.com/tohtsky/irspack/compare/v0.1.18...v0.1.19

- add two more data sets (Amazon music & CiteULike-a)
- add `ceil_n_heldout` option for split functions

* EDLAE recommender (https://github.com/tohtsky/irspack/pull/82)
* handle float32 score without casting to float64 (https://github.com/tohtsky/irspack/pull/83)
* Bug fix for NMF (https://github.com/tohtsky/irspack/pull/81)

We have also omitted the `numpy<=1.20` requirement (which was a hack to avoid a problem in optuna)

- https://github.com/tohtsky/irspack/pull/79 : Scikit-learn made optional & fix docstring
- https://github.com/tohtsky/irspack/pull/77 : fix for the config class