Refine good capture ordering by victim value

[KnightStackSys]

Stacks on top of move-ordering-see-split as a focused incremental move-ordering experiment.

Change

• keeps the existing SEE split intact
• only refines the SEE >= 0 capture bucket
• adds a modest captured-piece-value bonus (pieceValue(victim) * 32)
• SEE remains the primary tactical signal
• capture history still contributes as before
• losing captures remain below quiet-history moves
• en-passant resolves the actual captured pawn through Position::capturedSquare

Rationale

The first SEE split appears to be testing well. This follow-up tries to improve ordering within the already-good captures by preferring more valuable victims when SEE scores are otherwise close, without changing the broader move-stage ordering.

Scope

• based on move-ordering-see-split, not main
• only src/search.cpp changes
• no changes to PVS, LMR, LMP, SEE pruning, TT policy, history updates, MDP, aspiration windows, or evaluation

For clean testing, compare this branch directly against move-ordering-see-split.