Cache static evaluation with futility pruning

[KnightStackSys]

Extends the TT static-evaluation cache experiment with conservative futility pruning on top of current main.

TT/static-eval behavior

• caches raw static evaluation in the transposition table
• reuses cached TT eval in qsearch and negamax
• keeps raw staticEval separate from the TT-improved pruning eval
• TT searched scores only improve pruning eval when their bound is directionally compatible
• mate-range TT scores do not replace static evaluation
• keeps 16-byte TT entries and 64-byte clusters

Futility pruning

• applied after LMP and before SEE pruning
• non-PV nodes only
• disabled while in check
• quiet moves only
• skips TT moves and checking moves
• disabled in mate windows
• requires at least one legal move to have been searched
• FUTILITY_MAX_DEPTH = 2
• FUTILITY_MARGIN_PER_DEPTH = 200
• uses the TT-improved eval for the futility test

Existing NMP behavior preserved

• minimum depth 2
• partner-in-check guard
• dynamic reduction
• NULL_MOVE_DIVISOR = 190

For clean testing, SPRT this branch directly against current main.

[KnightStackSys]

Active test is running now