[DO NOT MERGE] Synthetic castling bug — CI mutation test

[ber4444]

⚠️ DO NOT MERGE — synthetic mutation / CI coverage probe

This PR intentionally breaks a chess rule to test whether CI catches the regression. It will be reverted immediately after the CI verdict is captured.

The bug injected

In chess-core/.../Move.kt, getCastlingMoves previously simulated the king's queenside transit through both c1 and d1 (resp. c8/d8) to verify neither square is attacked. This PR drops the d-square check:

-                // Simulate to check for check on d and c (b doesn't need to be unattacked)
-                val simD = allyPositions.toMutableList().apply { set(kingIndex, d) }
                 val simC = allyPositions.toMutableList().apply { set(kingIndex, c) }
-                if (!checkCheck(d, enemyPositions, enemyPieces, simD) &&
-                    !checkCheck(c, enemyPositions, enemyPieces, simC)) {
+                if (!checkCheck(c, enemyPositions, enemyPieces, simC)) {

That violates the FIDE castling rule: the king may not pass through a square that is attacked. Queenside castling is now legal even when d1/d8 is attacked.

Why this particular mutation

I deliberately picked a gap in the bespoke test suite:

• CastlingTest.kt does NOT cover this case. It tests the kingside king-path rule (f-file attacker) and the b1 exception, but never places an attacker on the d-file. → These unit tests still pass with the bug.
• The perft oracle (canonical node counts from chessprogramming.org) runs on castling-rich positions (Kiwipete, Position 4) and should diverge. → This is the safety net firing on a case the hand-written tests miss.

Local verification (before pushing)

Suite	Result
CastlingTest (bespoke unit tests)	✅ PASS — confirms the gap
PerftTest > kiwipete_depth_2	❌ FAIL — expected:<2039> but was:<2041> (+2 illegal castles)
PerftCanonicalGateTest > kiwipete_depth_3	❌ FAIL
PerftVsStockfishTest > canonical_positions_match_stockfish	❌ FAIL (Stockfish cross-check)

What CI should do

• ❌ check-kotlin (Linux) — :chess-core:check → :chess-core:desktopTest runs the perft rig → RED
• ❌ apple (macOS) — :chess-core:desktopTest --parallel → RED

If both jobs go red on PerftTest/PerftCanonicalGateTest, CI's regression coverage is confirmed. If any goes green, that's a real gap in the gate.

Outcome

Once CI finishes, this branch will be deleted and Move.kt restored to main. Do not merge.