Switching to a saner, and more conventional LMR scheme

[pkrisz99]

The previous implementation didn't make a whole lot of sense to me, and wanted to get rid of it for some time now. Finally, it passed non-regression.

Elo   | 1.94 +- 2.84 (95%)
SPRT  | 10.0+0.10s Threads=1 Hash=16MB
LLR   | 2.91 (-2.25, 2.89) [-4.00, 0.00]
Games | N: 15184 W: 3334 L: 3249 D: 8601
Penta | [59, 1739, 3904, 1838, 52]
https://zzzzz151.pythonanywhere.com/test/1091/

Renegade dev 1.1.26 Bench: 2600745