Share continuation history between threads
Passed 8th 5+0.05 https://tests.stockfishchess.org/tests/view/6a0bfdb46524d21ee79b879b
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 23472 W: 6103 L: 5823 D: 11546
Ptnml(0-2): 26, 2501, 6403, 2779, 27
Passed 8th 20+0.2 https://tests.stockfishchess.org/tests/view/6a0c87196524d21ee79b885c
LLR: 2.95 (-2.94,2.94) <0.50,2.50>
Total: 44692 W: 11640 L: 11319 D: 21733
Ptnml(0-2): 12, 4418, 13169, 4731, 16
Passed 16th 5+0.05 https://tests.stockfishchess.org/tests/view/6a20533f818cacc1db0ad32b
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 14720 W: 3910 L: 3642 D: 7168
Ptnml(0-2): 6, 1434, 4223, 1680, 17
Passed 64th 10+0.1 https://tests.stockfishchess.org/tests/view/6a20ae8e818cacc1db0ad369
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 34096 W: 8889 L: 8607 D: 16600
Ptnml(0-2): 4, 2974, 10808, 3260, 2
Continuation history is fixed size so there's a much more false sharing and a larger speed loss here at high thread counts, unfortunately.
vondele's machine, 4x70
```
==== master ====
Average (over 3): 294528128
==== shared-conthist ====
Average (over 3): 282364217 (~4% slowdown)
```
my machine, 8x32
```
Nodes/second : 243157385
Nodes/second : 228374554 (~6% slowdown)
```
Evidently it still gains at 64th, but a few followup ideas to try get the speed back:
- Add padding in `PieceToHistory` stats so there's less false sharing.
- Subdivide continuation history more finely than shared correction history.
- Scramble the `PieceToHistory` indexing so there's less false sharing
closes https://github.com/official-stockfish/Stockfish/pull/6905
No functional change