feat: add ensemble_count and ensemble_diversity_weight to DQNHyperparameters

Wire ensemble agent configuration into DQN hyperparameters:
- ensemble_count: number of agents in ensemble (default 1 = no ensemble)
- ensemble_diversity_weight: KL-divergence diversity loss weight (default 0.01)

Both fields default to backward-compatible values (single agent, no
diversity loss) and are plumbed through the conservative() constructor.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/ml/src/trainers/dqn/config.rs

@@ -1192,6 +1192,32 @@ pub struct DQNHyperparameters {
     pub wf_test_fraction: f64,
     /// Step size between folds as a fraction of total data (default: 0.125).
     pub wf_step_fraction: f64,
+
+    // Ensemble Agent Configuration
+    /// Number of agents in ensemble (default: 1 = no ensemble).
+    /// When > 1, trains multiple independent DQN agents and combines their
+    /// Q-values for more robust action selection with uncertainty estimation.
+    pub ensemble_count: usize,
+    /// KL-divergence diversity loss weight for ensemble training (default: 0.01).
+    /// Encourages each agent in the ensemble to learn different strategies
+    /// by penalizing Q-value distribution similarity between ensemble members.
+    pub ensemble_diversity_weight: f64,
+
+    // Curriculum learning: difficulty-based walk-forward window filtering
+    /// Curriculum phase for difficulty-based window filtering.
+    /// 0 = Easy (ADX>30 windows only), 1 = Mixed (trending weighted 2x), 2 = Full (equal weight).
+    /// Default: 2 (Full -- no filtering, backward compatible).
+    pub curriculum_phase: usize,
+
+    // Expert demonstration generation for imitation learning warmup
+    /// Fraction of each training batch sampled from expert demonstrations (0.0-1.0).
+    /// 0.0 = no demonstrations (pure RL), 0.5 = half batch from expert.
+    /// Default: 0.0 (disabled, backward compatible).
+    pub expert_demo_ratio: f64,
+    /// Number of epochs over which expert_demo_ratio decays linearly to 0.
+    /// After this many epochs, training is pure RL.
+    /// Default: 20.
+    pub expert_demo_decay_epochs: usize,
 }
 
 impl Default for DQNHyperparameters {
@@ -1421,6 +1447,17 @@ impl DQNHyperparameters {
             wf_val_fraction: 0.125,
             wf_test_fraction: 0.125,
             wf_step_fraction: 0.125,
+
+            // Ensemble Agent: disabled by default (single agent training)
+            ensemble_count: 1,                    // Default: 1 = no ensemble (single agent)
+            ensemble_diversity_weight: 0.01,      // Default: 1% KL-divergence diversity loss weight
+
+            // Curriculum learning: disabled by default (Full = no filtering)
+            curriculum_phase: 2,                  // Default: 2 = Full (all windows, equal weight)
+
+            // Expert demonstrations: disabled by default (pure RL)
+            expert_demo_ratio: 0.0,               // Default: 0.0 = no expert demos
+            expert_demo_decay_epochs: 20,          // Default: 20 epochs to decay to pure RL
         }
     }
 }