Add heuristic_balanced opponent_policy to interleaved scheduler

Mirrors the discard_only plumbing pattern. Uses HeuristicBot() (default
balanced params) and converts the bot's phase-local action to unified via
state.to_unified_action. Recursive (Cython) path was already supported
and is unchanged.

Tests: smoke run + accept/reject validators. All 59 tests pass.

src/coolrl_lost_cities/games/classic/deep_cfr/config.py

@@ -197,10 +197,16 @@ class TraversalConfig(StrictModel):
         if self.scheduler == "interleaved":
             if self.sampling_mode != "outcome":
                 raise ValueError("scheduler='interleaved' currently supports only outcome sampling")
-            if self.opponent_policy not in {"network", "average_strategy", "discard_only"}:
+            if self.opponent_policy not in {
+                "network",
+                "average_strategy",
+                "discard_only",
+                "heuristic_balanced",
+            }:
                 raise ValueError(
                     "scheduler='interleaved' currently supports only "
-                    "opponent_policy='network', 'average_strategy', or 'discard_only'"
+                    "opponent_policy='network', 'average_strategy', 'discard_only', "
+                    "or 'heuristic_balanced'"
                 )
             if self.cutoff_rollouts != 0 or self.cutoff_value_mode != "score_diff":
                 raise ValueError(

src/coolrl_lost_cities/games/classic/deep_cfr/interleaved_traversal.py

@@ -9,6 +9,7 @@ import numpy as np
 import torch
 
 from coolrl_lost_cities.games.classic.bots.discard_only import DiscardOnlyBot
+from coolrl_lost_cities.games.classic.bots.heuristic_py import HeuristicBot
 from coolrl_lost_cities.games.classic.deep_cfr.encoding import encode_info_state
 from coolrl_lost_cities.games.classic.deep_cfr.memory import TrainingSample
 from coolrl_lost_cities.games.classic.deep_cfr.traversal_stats import TraversalStats
@@ -381,6 +382,9 @@ class InterleavedContext:
         self._discard_only_bot: DiscardOnlyBot | None = (
             DiscardOnlyBot() if cfg.opponent_policy == "discard_only" else None
         )
+        self._heuristic_bot: HeuristicBot | None = (
+            HeuristicBot() if cfg.opponent_policy == "heuristic_balanced" else None
+        )
 
     def advance_until_policy(self, context_index: int) -> None:
         while not self.done and self.pending is None and self.stack:
@@ -482,6 +486,14 @@ class InterleavedContext:
             self.stack.append(FixedActionFrame(swapped_deck_index=swapped_deck_index))
             self.stack.append(EnterFrame(depth + 1))
             return
+        if player != self.traverser and self._heuristic_bot is not None:
+            phase_local_action = int(self._heuristic_bot.act(self.state))
+            action = int(self.state.to_unified_action(phase_local_action))
+            swapped_deck_index = self._sample_deck_draw_chance(action)
+            self.state.push_unified_action(action)
+            self.stack.append(FixedActionFrame(swapped_deck_index=swapped_deck_index))
+            self.stack.append(EnterFrame(depth + 1))
+            return
         info_state = encode_info_state(self.state, player, self.cfg.encoding)
         legal_mask = np.zeros(self.cfg.action_size, dtype=bool)
         legal_mask[legal_actions] = True