feat(dqn): wire QR-DQN QuantileNetwork through hyperopt to training loop

Add use_qr_dqn, num_quantiles, qr_kappa fields to DQNHyperparameters
and pass them from DQNParams (hyperopt) through to DQNConfig (agent),
replacing hardcoded IQN values in the trainer.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ml/src/hyperopt/adapters/dqn.rs

@@ -2239,6 +2239,11 @@ impl HyperparameterOptimizable for DQNTrainer {
             noisy_sigma_initial: 0.6,            // Default: 60% initial noise
             noisy_sigma_final: 0.4,              // Default: 40% final noise
             noisy_sigma_anneal_steps: 10000,     // Default: 10K steps for annealing
+
+            // QR-DQN (replaces disabled C51)
+            use_qr_dqn: params.use_qr_dqn,
+            num_quantiles: params.num_quantiles,
+            qr_kappa: params.qr_kappa,
         };
 
         let data_path_str = self

ml/src/trainers/dqn/config.rs

@@ -579,6 +579,14 @@ pub struct DQNHyperparameters {
     /// Prevents overfitting by penalizing large weights via L2 regularization.
     /// Recommended: 1e-4 for standard training, 1e-3 for aggressive regularization.
     pub weight_decay: f64,
+
+    // QR-DQN (Quantile Regression DQN) — replaces disabled C51
+    /// Enable QR-DQN (IQN) for distributional RL
+    pub use_qr_dqn: bool,
+    /// Number of quantiles for QR-DQN (32-200)
+    pub num_quantiles: usize,
+    /// Kappa for quantile Huber loss (0.5-2.0)
+    pub qr_kappa: f64,
 }
 
 impl Default for DQNHyperparameters {
@@ -748,6 +756,11 @@ impl DQNHyperparameters {
 
             // WAVE 30: L2 Weight Decay
             weight_decay: 1e-4,                   // Default: 0.0001 (standard regularization strength)
+
+            // QR-DQN (replaces disabled C51)
+            use_qr_dqn: true,                     // Default: enabled (IQN distributional RL)
+            num_quantiles: 32,                    // Default: 32 quantiles
+            qr_kappa: 1.0,                        // Default: 1.0 (standard quantile Huber loss)
         }
     }
 }

ml/src/trainers/dqn/trainer.rs

@@ -350,10 +350,10 @@ impl DQNTrainer {
 
             use_cql: true,
             cql_alpha: 1.0,
-            use_iqn: true,
-            iqn_num_quantiles: 64,
-            iqn_kappa: 1.0,
-            iqn_embedding_dim: 64,
+            use_iqn: hyperparams.use_qr_dqn,                  // Controlled by hyperopt
+            iqn_num_quantiles: hyperparams.num_quantiles,       // Controlled by hyperopt
+            iqn_kappa: hyperparams.qr_kappa as f32,            // Controlled by hyperopt (f64→f32)
+            iqn_embedding_dim: 64,                              // Fixed (not in search space)
             use_cvar_action_selection: false,
             cvar_alpha: 0.05,
         };