[FEATURE] Add eval margin scaling for NMP

[ptsouchlos]

Now that we can track evaluation for each ply, we can add eval scaling to NMP.

[ptsouchlos]

I gave this a try without verification (#358) but this failed:

Elo   | -4.42 +- 5.13 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | -2.97 (-2.94, 2.94) [0.00, 5.00]
Games | N: 7854 W: 2133 L: 2233 D: 3488
Penta | [219, 967, 1611, 955, 175]

https://openbench.nocturn9x.space/test/7114/

I also tried adding NMP verification on it's own (#363), which also failed:

Elo   | -2.88 +- 2.27 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | -3.03 (-2.94, 2.94) [-5.00, 0.00]
Games | N: 40044 W: 10700 L: 11032 D: 18312
Penta | [1057, 4922, 8288, 4806, 949]

https://openbench.nocturn9x.space/test/7117/

Then I tried building eval margin and making NMP more aggressive with verification, and these both failed (#364 , #365):

Elo   | -5.09 +- 5.42 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | -2.95 (-2.94, 2.94) [0.00, 5.00]
Games | N: 7038 W: 1883 L: 1986 D: 3169
Penta | [185, 898, 1437, 833, 166]

https://openbench.nocturn9x.space/test/7133/

Elo   | -0.19 +- 2.83 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | -2.96 (-2.94, 2.94) [0.00, 5.00]
Games | N: 25656 W: 6910 L: 6924 D: 11822
Penta | [641, 3130, 5294, 3128, 635]

https://openbench.nocturn9x.space/test/7134/

It could just be that we don't have other features implemented (yet) to be able to overcome the performance penalty of the verification search.