feat(snake): add adaptive adversarial search

- Share duel search contexts and transpositions across candidate moves.
- Add aspiration windows, principal variation ordering, and body caches.
- Model simultaneous multiplayer responses with a compact beam rollout.
- Adapt search depth and response breadth to the remaining deadline.
- Add a deterministic arena benchmark with optional JSON reporting.
- Expose search metrics, document benchmarking, and bump Prism to 1.2.0.

README.md

@@ -68,9 +68,12 @@ Allowed values: `safe`, `balanced`, `aggressive`.
 
 ## PrismBattleSnake_GPT_5_6_Sol
 `PrismBattleSnake_GPT_5_6_Sol` is a separate snake that keeps Apex's strategy while
-accelerating hot spatial operations with a Python-integer bitboard engine. Its
-filename, class, and registry key include the model name, while its public
-Battlesnake API name remains `PrismBattleSnake`.
+accelerating hot spatial operations with a Python-integer bitboard engine. It
+also shares duel transpositions across candidate moves, uses principal-variation
+ordering and aspiration windows, and runs a compact adversarial multiplayer
+rollout with simultaneous enemy responses. Its filename, class, and registry
+key include the model name, while its public Battlesnake API name remains
+`PrismBattleSnake`.
 
 Run it with:
 ```sh
@@ -87,6 +90,17 @@ python scripts/benchmark_snakes_from_db.py \
 The benchmark opens SQLite read-only and reports mean, median, p95, and maximum
 move latency. Increase `--samples` for a broader but slower comparison.
 
+Run the deterministic CI-friendly arena benchmark without a gameplay database:
+```sh
+just bench-snake-arena positions=100
+```
+
+It reports latency, reached minimax depth, and move disagreements between Apex
+and Prism. Add `output=data/arena-report.json` to save a machine-readable report.
+For representative strategy evaluation, provide recorded positions to
+`scripts/benchmark_snake_arena.py --database /path/to/gameplay.sqlite3`, then run
+paired seeded games with the local Battlesnake CLI to measure win rate.
+
 ## Compact gameplay database
 New gameplay turns use normalized storage: the turn row stores food, hazards,
 move, and thinking data once; snake identity is stored once per game in