cleanup(fflag): delete dead enable_gradient_vaccine chain — [FFLAG-012]

Same dead-flag chain as FFLAG-011: GpuDqnTrainConfig.enable_gradient_vaccine
→ local vaccine_enabled → GpuDqnTrainer.enable_gradient_vaccine field.
Zero readers for self.enable_gradient_vaccine — the vaccine kernels
(vaccine_dot_kernel, vaccine_project_kernel) run unconditionally. Deleted
field + struct field + local + setter at fused_training, 3 sites total.

crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs

@@ -310,10 +310,6 @@ pub struct GpuDqnTrainConfig {
     /// #20 Fraction of weights to prune (0.7 = remove 70% smallest).
     pub pruning_fraction: f32,
 
-    /// #32 Enable gradient vaccine. When true, projects out overfitting gradient
-    /// components using a held-out validation batch.
-    pub enable_gradient_vaccine: bool,
-
     /// #31 Temporal causal bottleneck dimension. 0 = disabled (no bottleneck).
     /// 2 = maximum compression (14 market features → 2 abstract dimensions).
     pub bottleneck_dim: usize,
@@ -365,7 +361,6 @@ impl Default for GpuDqnTrainConfig {
             pruning_fraction: 0.7,
             causal_weight: 0.1,
             causal_intervention_interval: 50,
-            enable_gradient_vaccine: true,     // always on
             bottleneck_dim: 16,
             market_dim: 42, // Default: 42 base features. Overridden to 50 when OFI (MBP-10) enabled.
         }
@@ -1141,8 +1136,6 @@ pub struct GpuDqnTrainer {
     vaccine_dot_kernel: CudaFunction,
     vaccine_dot_finalize_kernel: CudaFunction,
     vaccine_project_kernel: CudaFunction,
-    /// #32 Whether gradient vaccine is enabled.
-    enable_gradient_vaccine: bool,
 
     /// #31 Bottleneck hidden activation [B, bottleneck_dim] f32. None when bottleneck_dim=0.
     bn_hidden_buf: CudaSlice<f32>,
@@ -6388,7 +6381,6 @@ impl GpuDqnTrainer {
         let dd_weight = config.asymmetric_dd_weight;
         let ens_weight = config.ensemble_disagreement_penalty;
         let sd_prob = config.stochastic_depth_prob;
-        let vaccine_enabled = config.enable_gradient_vaccine;
         let prune_ep = config.pruning_epoch;
         let prune_frac = config.pruning_fraction;
         let bn_dim = config.bottleneck_dim;
@@ -7473,7 +7465,6 @@ impl GpuDqnTrainer {
             vaccine_dot_finalize_kernel: vaccine_dot_finalize,
             vaccine_block_results,
             vaccine_project_kernel,
-            enable_gradient_vaccine: vaccine_enabled,
             bn_hidden_buf,
             bn_concat_buf,
             bn_d_concat_buf,

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

@@ -405,7 +405,6 @@ impl FusedTrainingCtx {
             pruning_fraction: hyperparams.pruning_fraction as f32,
             causal_weight: hyperparams.causal_weight,
             causal_intervention_interval: hyperparams.causal_intervention_interval,
-            enable_gradient_vaccine: true,
             bottleneck_dim: hyperparams.bottleneck_dim,
             market_dim: 42, // Always 42 base market features — OFI features bypass bottleneck via portfolio_dim
         };