Implement the C&P algorithm from arXiv:1902.08036

[Naereen]

Cf. #169, #183.
The recent paper [Multi-Player Bandits: The Adversarial Case, by Pragnya Alatur, Kfir Y. Levy, Andreas Krause, arXiv:1902.08036] proposes an efficient algorithm for the adversarial case of multi-player MAB.
Their algorithm seems to also be efficient for piece-wise stationary problem (and not only adversarial problems)

• I should read it carefully,
• And implement in SMPyBandits their algorithms,
• To do my own comparison against RandTopM and MCTopM, and Selfish, on piecewise stationary problems with a small number of changes,
• And check and verify their claims. (or disprove them?),

Note that their review of the current state of the art is not complete, they basically missed all the works between Musical Chair (ICML 2016) and Game-of-Thrones (NeurIPS 2018), so they missed our work on MCTopM, and the subsequent works of Avner & Mannor, August 2018 #139, Lugosi & Mehrabian, August 2018 #141, Bourse & Perchet, September 2018 #145.