feat: refactor AlgorithmicEvaluator to stateless, add game engine module and holdem_sim CLI

- Refactor AlgorithmicEvaluator from stateful Arc<Evaluator> wrapper to
  pure stateless evaluator with zero-allocation construction
- Extract shared evaluation logic into pub(crate) methods, eliminating
  ~220 lines of duplicated code from Evaluator
- Add game module: table, hand runner, tournament, bots, config, stats
- Add holdem_sim CLI binary crate for running simulations
- Convert MathEvaluator to thread-safe singleton with Arc<Inner> pattern
- Replace heap-allocated combination generation with const-generics
  stack-based approach in AlgorithmicEvaluator::evaluate_n_card_hand
- Add thiserror 2 dependency for GameError derive
- Update docs: README, ACHIEVEMENTS, ROADMAP

holdem_core/Cargo.toml

@@ -25,6 +25,7 @@ once_cell = "1.19"
 chrono = { version = "0.4", features = ["serde"] }
 indicatif = "0.17"
 log = "0.4"
+thiserror = "2"
 
 [dev-dependencies]
 proptest = "1.0"