feat(ml): add KAN architecture + TLOB/KAN trainable/hyperopt adapters

Phase 2-3 of ensemble expansion:
- KAN (Kolmogorov-Arnold Network): B-spline basis, layer, network, trainable adapter
- TLOB UnifiedTrainable adapter with 3D input support (batch, seq, features)
- Hyperopt adapters for both KAN and TLOB (ParameterSpace + metrics)
- ModelType::KAN variant registered in common, coordinator, lib.rs
- 44 new tests, all passing, zero warnings

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

ml/src/integration/coordinator.rs

@@ -471,6 +471,10 @@ impl EnsembleCoordinator {
                 // Ensemble methods - use weighted combination approach
                 self.deep_q_prediction(input_features, model.weight * 1.2) // Enhanced prediction for ensemble
             },
+            ModelType::KAN => {
+                // Kolmogorov-Arnold Network prediction
+                self.deep_q_prediction(input_features, model.weight)
+            },
         };
 
         let latency_us = start_time.elapsed().as_micros() as u64;
@@ -882,6 +886,7 @@ impl EnsembleCoordinator {
             ModelType::PPO => 0.83,         // Good confidence for policy optimization
             ModelType::Transformer => 0.88, // High confidence for sequence modeling
             ModelType::Ensemble => 0.92,    // Highest confidence for ensemble methods
+            ModelType::KAN => 0.86,         // Good confidence for learned activation functions
         };
 
         // Adjust confidence based on feature quality