Refactor of the Game and Strategy API

[gavento]

Games now defiend only in Game descendant. Game state is always GameState, this includes new-observetion style observations (with Observation to distinguish own action and others) and Active player info objects.

Takes a lot of inspiration from https://github.com/gavento/gtcogs.

Also inludes polish and improvements such as:

• MCCFR gradual burn-in for better performance
• MCCFR supports updating a subset of players
• approximate best response via MCCFR
• MCCFR has nice progress bars with tqdm
• node limits for best response
• Removed gamegym.distribution in favor of simple utils.DIstribution
• Removed MCCFR persist, added utils.store/utils.load

TODOs:

• Update one card poker and its tests
• Update all value learning and its tests

[coveralls]

Pull Request Test Coverage Report for Build 31

• 472 of 524 (90.08%) changed or added relevant lines in 12 files are covered.
• 123 unchanged lines in 6 files lost coverage.
• Overall coverage decreased (-6.6%) to 63.028%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
gamegym/games/matrix.py	18	22	81.82%
gamegym/algorithms/infosets.py	31	40	77.5%
gamegym/strategy.py	25	34	73.53%
gamegym/game.py	118	128	92.19%
gamegym/utils.py	32	52	61.54%

Files with Coverage Reduction	New Missed Lines	%
gamegym/algorithms/infosets.py	1	87.21%
gamegym/games/matrix.py	4	83.67%
gamegym/games/goofspiel.py	11	84.51%
gamegym/value_learning/valuestore.py	28	0.0%
gamegym/value_learning/valuesgd.py	36	0.0%
gamegym/games/onecardpoker.py	43	23.21%

Totals
Change from base Build 26:	-6.6%
Covered Lines:	716
Relevant Lines:	1136

---

💛 - Coveralls

[gavento]

Replaced by #16