Rarity: Stockfish 104 Elo patch

[Jouni]

Improve performance on NUMA systems.

Passed 60+1 @ 256t 16000MB hash: https://tests.stockfishchess.org/tests/ ... d5e27db0d8
```
LLR: 2.95 (-2.94,2.94) <0.00,10.00>
Total: 278 W: 110 L: 29 D: 139 Elo +104.25
Ptnml(0-2): 0, 1, 56, 82, 0
```

Also no regression in 1 or 7 cores .

[CornfedForever]

Improve performance on NUMA systems.

Passed 60+1 @ 256t 16000MB hash: https://tests.stockfishchess.org/tests/ ... d5e27db0d8
```
LLR: 2.95 (-2.94,2.94) <0.00,10.00>
Total: 278 W: 110 L: 29 D: 139 Elo +104.25
Ptnml(0-2): 0, 1, 56, 82, 0
```

Also no regression in 1 or 7 cores .

Saw that...didn't understand it.

[Werewolf]

Improve performance on NUMA systems.

Passed 60+1 @ 256t 16000MB hash: https://tests.stockfishchess.org/tests/ ... d5e27db0d8
```
LLR: 2.95 (-2.94,2.94) <0.00,10.00>
Total: 278 W: 110 L: 29 D: 139 Elo +104.25
Ptnml(0-2): 0, 1, 56, 82, 0
```

Also no regression in 1 or 7 cores .

Do you have to set NUMA up for SF, or does SF sort it out for you? I read somewhere choosing the number of threads before setting hash makes a big difference.