fix: add missing iat_cv arm in feature extractor

The v10 model's ae_feature_names includes iat_cv (feature index 30),
but PrecomputedStats::get() had no matching arm. This caused the
feature to fall through to the default 0.0, which after AE scaling
became (0-μ)/σ — a constant bias on every inference since v10 shipped.

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

net-guardia/src/core/ml/feature_extractor.rs

@@ -379,6 +379,9 @@ impl PrecomputedStats {
             "pkt_len_variance" => self.all_len_std * self.all_len_std,
             "fwd_iat_skewness" => self.fwd_iat_skewness,
 
+            // Model lists this in ae_feature_names but had no arm — was receiving post-scaler (0-μ)/σ.
+            "iat_cv" => safe_div(self.flow_iat_std, self.flow_iat_mean),
+
             _ => 0.0,
         }
     }