No major features added, mostly bug fixes and refactoring. Details include:
- Correction histories were completely broken before, but did not affect the evaluation due to the improperly implemented
CORRECTION_GRAINscaling it down to values small enough to not have an impact. This is now fixed by tracking keys separately. - Fixed the broken PV nodes replacement condition in TT.
- Split previously huge inline headers into separate headers and source files to make clangd happy.
- Experimented with SPSA tuning with random time controls to avoid overfitting the parameters.
- And many other minor fixes.
Approximately 54 Elo stronger than v1.5 in STC and around the same for LTC.