[2/6] Fix transposition table with proper bounds

[luccabb]

Summary

• Add Bound enum (EXACT, LOWER_BOUND, UPPER_BOUND) for correct TT behavior
• Use Zobrist hash as cache key instead of FEN string (faster)
• Store bound type and depth with each cache entry
• Fix null move pruning condition (was missing null_move parameter check)
• Update parallel engines to use new cache format

Details

The transposition table now correctly stores and uses bound information:

• EXACT: Score is within the alpha-beta window, can be used directly
• LOWER_BOUND: Node failed high (score >= beta), true score is at least this value
• UPPER_BOUND: Node failed low (score <= alpha), true score is at most this value

This prevents incorrect cutoffs and score usage that can cause search instability.

Parallel Engine Updates

All parallel engines (lazy_smp, l1p, l2p) updated to:

• Use Zobrist hash for cache key lookup
• Use context managers for proper Pool/Manager cleanup
• Fix score negation in l1p (opponent perspective -> our perspective)

Test plan

• All 64 unit tests pass
• Verified correct bound handling in search

---

🤖 Generated with Claude Code

[github-actions]

🔬 Stockfish Benchmark Results

vs Stockfish Skill Level 3

Metric	Wins	Losses	Draws	Total	Win %
Overall	10	88	2	100	10.0%
As White	7	42	1	50	14.0%
As Black	3	46	1	50	6.0%

Non-checkmate endings:

• Draw by 3-fold repetition: 2

vs Stockfish Skill Level 4

Metric	Wins	Losses	Draws	Total	Win %
Overall	7	91	2	100	7.0%
As White	3	46	1	50	6.0%
As Black	4	45	1	50	8.0%

Non-checkmate endings:

• Draw by insufficient mating material: 1
• Draw by 3-fold repetition: 1

vs Stockfish Skill Level 5

Metric	Wins	Losses	Draws	Total	Win %
Overall	2	97	1	100	2.0%
As White	2	47	1	50	4.0%
As Black	0	50	0	50	0%

Non-checkmate endings:

• Draw by 3-fold repetition: 1

Configuration

• 5 chunks × 20 rounds × 3 skill levels = 300 total games
• Each opening played with colors reversed (-repeat) for fairness
• Moonfish: 60s per move
• Stockfish: 60+5 time control