Total Elo gain estimate: ~200 Elo
- improved time management for shorter time controls
- lazy nnue updates
- fused nnue updates
- feature set changed from halfKA to halfKA_hm
- nnue output buckets
- 2 cycles of (generate new training data -> retrain network)
- switch from relative history heuristic to normal history heuristic
- improved SEE implementation
- SEE pruning in qsearch
- only do attempt nmp when staticEval >= beta
- drop remaining depth requirement for late move reductions to >= 2