Pygmtools

* Support CUDA ops for MGM
* Add "device" argument for ``from_numpy``
* Fix testing script for GAMGM

**Full Changelog**: https://github.com/Thinklab-SJTU/pygmtools/compare/0.2.2...0.2.3

* Add multi-graph matching algorithms (cao, mgm-floyd, gamgm). Now ``pytorch`` backend is supported.
* Add CI support for multi-graph matching
* Bug fixes

**Full Changelog**: https://github.com/Thinklab-SJTU/pygmtools/compare/0.2.1...0.2.2

* Add CI support
* Bug fix: https://github.com/Thinklab-SJTU/pygmtools/issues/1

**Full Changelog**: https://github.com/Thinklab-SJTU/pygmtools/compare/0.2.0...0.2.1

We include:
+ Benchmark on 5 datasets to handle dataset downloading/processing and evaluation.
+ You may also implement other datasets by this API.
+ Classic graph matching solvers on pytorch, numpy backend.