config: enable position entropy (0.02) + add hyperopt search range

Anti-Flat-collapse: rewards exposure diversity via per-episode
position visit histogram entropy bonus.

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

config/training/dqn-hyperopt.toml

@@ -78,6 +78,7 @@ w_idle = [0.0, 0.1]
 dd_threshold = [0.005, 0.03]            # HFT: tight drawdown tolerance (0.5%-3%)
 loss_aversion = [1.0, 1.0]
 time_decay_rate = [0.0001, 0.005]
+position_entropy_weight = [0.01, 0.1]
 
 # Trade conviction filter
 q_gap_threshold = [0.0, 0.5]          # 0.0=trade every bar, 0.5=high conviction only

config/training/dqn-localdev.toml

@@ -94,7 +94,7 @@ stochastic_depth_prob = 0.2
 asymmetric_dd_weight = 0.5
 time_reversal_mod = 5
 regret_blend = 0.3
-position_entropy_weight = 0.01
+position_entropy_weight = 0.02
 trade_clustering_penalty = 0.05
 ensemble_disagreement_penalty = 0.3
 mixup_alpha = 0.2

config/training/dqn-production.toml

@@ -105,7 +105,7 @@ stochastic_depth_prob = 0.2
 asymmetric_dd_weight = 0.5
 time_reversal_mod = 5
 regret_blend = 0.3
-position_entropy_weight = 0.01
+position_entropy_weight = 0.02
 trade_clustering_penalty = 0.05
 ensemble_disagreement_penalty = 0.3
 mixup_alpha = 0.2

config/training/dqn-smoketest.toml

@@ -96,7 +96,7 @@ trade_clustering_penalty = 0.02 # 0.02 vs production 0.05
 ensemble_disagreement_penalty = 0.1  # 0.1 vs production 0.3
 regret_blend = 0.15            # 0.15 vs production 0.3
 time_reversal_mod = 10         # 10% reversed vs production 20%
-position_entropy_weight = 0.005 # Half of production
+position_entropy_weight = 0.02
 causal_weight = 0.05           # Half of production
 causal_intervention_interval = 50  # Less frequent than production (10)
 bottleneck_dim = 0             # Disabled for smoketest (tiny network can't afford bottleneck)