cleanup(fflag): delete dead enable_action_masking chain — [FFLAG-013b]

The chain was: GpuDqnTrainer.enable_action_masking field (mod.rs:234) →
set via local `let enable_action_masking = true;` at constructor.rs:380
→ passed into GpuExperienceCollectorConfig.enable_action_masking at
training_loop.rs:1263. Zero conditional readers anywhere in cuda_pipeline
— the GPU kernel always filters invalid actions. Comment at
constructor.rs:379 says "action masking and entropy regularization
always active", confirming the flag is vestigial.

Deleted all 4 sites. Per user directive "no deferred tasks, solve
properly" — previously deferred as FFLAG-013b, now resolved.

crates/ml/src/cuda_pipeline/gpu_experience_collector.rs

@@ -237,8 +237,6 @@ pub struct ExperienceCollectorConfig {
     pub dsr_eta: f32,
     /// N-step returns lookahead (1 = standard TD, 3-5 typical for Rainbow DQN)
     pub n_steps: i32,
-    /// Enable action masking: filters invalid exposure actions in the GPU kernel.
-    pub enable_action_masking: bool,
     /// Holding cost rate per bar per unit position. Continuous inventory penalty.
     pub holding_cost_rate: f32,
     /// Bars threshold for churn penalty. Trades within this window get graduated penalty.
@@ -370,7 +368,6 @@ impl Default for ExperienceCollectorConfig {
             q_gap_threshold: 0.0,
             dsr_eta: 0.01,
             n_steps: 1,
-            enable_action_masking: false,
             holding_cost_rate: 0.00000052,  // ~0.52 bps / 100 bars
             churn_threshold: 10.0,
             churn_penalty_scale: 0.001,

crates/ml/src/trainers/dqn/trainer/constructor.rs

@@ -377,7 +377,6 @@ impl DQNTrainer {
         };
 
         // Wave 16 Portfolio Features: action masking and entropy regularization always active
-        let enable_action_masking = true;
         let max_position = hyperparams.max_position_absolute; // BLOCKER #2: Use hyperopt-tunable position limit
 
         // Entropy regularization: SAC-style computed directly on Q-values in DQN::compute_loss_internal
@@ -572,8 +571,7 @@ impl DQNTrainer {
             training_sharpe_ema: 0.0,
             training_sharpe_ema_initialized: false,
 
-            // Wave 16 Portfolio Features
-            enable_action_masking,
+            // Wave 16 Portfolio Features (action masking always active)
             max_position,
             current_epoch: 0,
             multi_asset_portfolio,

crates/ml/src/trainers/dqn/trainer/mod.rs

@@ -229,9 +229,7 @@ pub struct DQNTrainer {
     /// Whether training_sharpe_ema has been initialized.
     pub(crate) training_sharpe_ema_initialized: bool,
 
-    // Wave 16 Portfolio Features
-    /// Enable action masking (filters invalid actions before Q-value computation)
-    pub enable_action_masking: bool,
+    // Wave 16 Portfolio Features (action masking is always active).
     /// Maximum position size for action masking (default: 2.0)
     pub max_position: f64,
     /// Current epoch (for Q-gap warm-up ramp)

crates/ml/src/trainers/dqn/trainer/training_loop.rs

@@ -1260,7 +1260,6 @@ impl DQNTrainer {
             epsilon,
             gamma: self.hyperparams.gamma as f32,
             max_position: self.max_position as f32,
-            enable_action_masking: self.enable_action_masking,
             curiosity_scale: 1.0,
             order_credit_weight: self.hyperparams.order_credit_weight as f32,
             risk_efficiency_weight: self.hyperparams.risk_efficiency_weight as f32,