Add multithreading to PZChessBot

[kevlu8]

• This PR implements Lazy SMP as a way to perform multi-threaded search in PZChessBot.

Benchmarks:

• 1 thread: 2.5Mnps
• 2 threads: 5.4Mnps
• 4 threads: 9.6Mnps
• 8 threads: 17.2Mnps

Current issues:

• Elo gains are lower than expected
• Code is very messy

Notes:

• MultiPV is removed as it was broken in prior commits
• HCE is removed as it unnecessarily bloats code complexity
• Some UCI outputs have been removed (e.g. seldepth) but will be added back shortly

TODO:

• Refactor threading code
• Allocate memory for new threads as we go instead of initializing a global for MAX_THREADS (slow!)
• Start search on separate thread to allow for uci stop command

[kevlu8]

SPRT for always-replace:

Elo   | 13.44 +- 7.44 (95%)
SPRT  | 8.0+0.08s Threads=2 Hash=32MB
LLR   | 3.01 (-2.25, 2.89) [0.00, 5.00]
Games | N: 4732 W: 1786 L: 1603 D: 1343
Penta | [204, 388, 1038, 493, 243]

https://sscg13.pythonanywhere.com/test/1593/

[kevlu8]

Preliminary test of SMP vs main:

Elo   | -15.44 +- 10.98 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=32MB
LLR   | -0.98 (-2.20, 2.20) [-10.00, -5.00]
Games | N: 1126 W: 228 L: 278 D: 620
Penta | [12, 146, 293, 104, 8]

https://sscg13.pythonanywhere.com/test/1607/

SMP seems to be clearly weaker. This may be due to several changes that were untested on main, such as:

• Changing TT scheme to always-replace
• Copying TT entry at every node

[kevlu8]

New test:

Elo   | -3.99 +- 2.63 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=32MB
LLR   | -2.36 (-2.25, 2.89) [-8.00, -2.00]
Games | N: 18970 W: 4258 L: 4476 D: 10236
Penta | [102, 2404, 4679, 2210, 90]

https://sscg13.pythonanywhere.com/test/1613/