feat(ml): add 4 innovation kernels - streaming anomaly and explainability (98 total)

Add 4 new ML kernels for real-time anomaly detection and model explainability:

Streaming Anomaly Detection:
- StreamingIsolationForest: Online anomaly detection with sliding window,
  incremental forest updates, and configurable rebuild intervals.
- AdaptiveThreshold: Self-adjusting thresholds with Welford's online
  statistics, concept drift detection, and optional ground truth feedback.

Explainability:
- SHAPValues: Kernel SHAP approximation for model-agnostic feature
  explanations. Supports both sampling and weighted regression methods.
- FeatureImportance: Permutation-based importance with multiple metrics
  (accuracy, MSE, MAE, R2) and ranking output.

All 50 ML tests pass. Kernel count: 8 → 12 in ML crate, 98 total.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: mivertowski

README.md

@@ -26,7 +26,7 @@ Most kernels support both modes. Choose based on your latency requirements.
 | Domain | Crate | Kernels |
 |--------|-------|---------|
 | Graph Analytics | `rustkernel-graph` | 26 |
-| Statistical ML | `rustkernel-ml` | 8 |
+| Statistical ML | `rustkernel-ml` | 12 |
 | Compliance | `rustkernel-compliance` | 11 |
 | Temporal Analysis | `rustkernel-temporal` | 7 |
 | Risk Analytics | `rustkernel-risk` | 5 |
@@ -94,7 +94,7 @@ crates/
 
 ## Status
 
-The port is functionally complete. All 94 kernels have been implemented with both BatchKernel and RingKernelHandler traits. K2K (kernel-to-kernel) messaging is in place for cross-kernel coordination patterns.
+The port is functionally complete. All 98 kernels have been implemented with both BatchKernel and RingKernelHandler traits. K2K (kernel-to-kernel) messaging is in place for cross-kernel coordination patterns.
 
 Test coverage exists for all domains. Some edge cases in the behavioral analytics causal graph module remain flaky.