Molybdenum v4.0

This release features an Elo gain of around 35:

STC:

Elo   | 61.08 +- 16.03 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=8MB
Games | N: 1000 W: 359 L: 185 D: 456
Penta | [5, 74, 198, 188, 35]
https://aytchell.eu.pythonanywhere.com/test/492/

LTC:

Elo   | 36.62 +- 15.11 (95%)
Conf  | 40.0+0.40s Threads=1 Hash=64MB
Games | N: 1000 W: 297 L: 192 D: 511
Penta | [2, 77, 252, 152, 17]
https://aytchell.eu.pythonanywhere.com/test/490/

Elo gains:

• Search Improvements:

  • Dont extend all moves when singular extending: #81
  • Better killer handling at root: #82
  • Research lmr fail highs: #85
  • Do deeper but worse: #89

• NNUE improvements:

  • Train on an additional 100m fens: #80
  • Net trained on 0.5 wdl: #84

This release also features a few non functional speedups and the Introduction of Lazy SMP (#93), meaning Molybdenum can now be run with more than one thread.

The provided binaries use avx2, bmi2... (v3) / sse3, popcnt.. (v2) optimization respectively. The avx2bmi2 binary is faster so it should be preferred if your CPU supports these instructions sets. Note that even ryzens before zen 3 should be able to use the bmi2 binaries without slowdown, as they contain no pext / pdep instructions