feat(search): killer/history ordering, null-move pruning, LMR

[cjunius]

What

Adds three classic search heuristics on top of the transposition table and Lazy SMP that landed in #8.

Killer moves + history heuristic

• Two killer slots per ply for quiet moves that caused a beta cut-off.
• [side][from][to] history table, += depth² per cut-off, clamped at 1<<22.
• orderMoves now ranks: TT move → promotions → MVV-LVA captures → killers → quiets by history score.
• Both tables live on the per-worker searcher, so Lazy SMP stays lock-free.

Null-move pruning

• R = 2, R = 3 from depth 6.
• Guards: not in check, depth >= 3, non-mate beta, side-to-move has non-pawn material (zugzwang), no consecutive nulls, static eval ≥ beta.
• The pass position is rebuilt through FEN — dragontoothmg keeps the Zobrist hash and en-passant square in unexported fields, so a plain struct copy would feed a stale hash into the shared TT.

Late move reductions

• Late quiet moves that neither give nor evade check are searched 1–2 ply shallower.
• A reduced search that beats alpha is re-searched at full depth, so no score is lost.

negamax is split into outOfTime / terminalScore / tryNullMove / searchMove to stay within the gocyclo budget.

Impact

Single-threaded, Hash 128, from the start position:

depth	nodes (TT only, was)	nodes (now)	time (now)
8	~5.9M	~167k	0.07s
10	—	~548k	0.23s
12	—	~2.29M	1.13s

The engine now reaches depth 12 from the opening in roughly the wall-clock depth 8 used to take. Node rate is ~unchanged (~2.5M nps).

Tests

• New engine internal tests: nullMoveBoard, hasNonPawnMaterial, isQuiet, recordCutoff (killer shift + history cap), orderMoves ranking, tryNullMove guard matrix, board-unmodified invariant.
• New external search tests: mate-with-pruning-active, deep-depth completion, tactical parity with the pre-pruning result.
• Broader UCI tests: setoption Hash/Threads (incl. clamping and bad values), ucinewgame reset, clock-budget go, malformed position inputs.
• New cmd/gochess/main_test.go covering runBench, runPerft, dispatch, and the UCI entry point.

Coverage: engine 90.4%, uci 95.8%, cmd 88.4% — 91.5% overall.

golangci-lint (v2.13.2), gofumpt, go vet, go test -race ./..., and the full perft suite all pass.

Docs

README, docs/architecture.md, docs/design.md, docs/performance.md, docs/engine-strength.md, and CHANGELOG.md updated for the new heuristics and refreshed benchmark tables. The strength estimate is revised upward (~1600 → ~2000 CCRL-blitz, still an estimate pending SPRT).

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 95.69892% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/engine/search.go	95.69%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!