Retrained network with DFRC data

[liamt19]

Added 0.8b DFRC positions, and retrained from superbatch 315 using cosine annealing.

Might be slightly better at standard:

Elo   | 2.23 +- 2.90 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | 2.23 (-2.94, 2.94) [-3.00, 0.00]
Games | N: 15904 W: 4000 L: 3898 D: 8006
Penta | [107, 1814, 4006, 1920, 105]
http://somelizard.pythonanywhere.com/test/1058/

And significantly better at DFRC:

Elo   | 26.55 +- 7.99 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | 2.95 (-2.94, 2.94) [0.00, 3.00]
Games | N: 3108 W: 911 L: 674 D: 1523
Penta | [67, 264, 678, 455, 90]
http://somelizard.pythonanywhere.com/test/1059/