feat(mincut): complete j-tree coordinator integration

- coordinator.rs: Fixed to work with JTreeHierarchy
  - Lazy initialization pattern with ensure_built()
  - EscalationPolicy enum (Never, Always, LowConfidence, etc.)
  - TierMetrics for usage tracking
  - 14 coordinator-specific tests passing

- mod.rs: Export coordinator types
- benchmark.rs: Minor refinements
- parallel.rs: Minor refinements

All 50 jtree tests now passing.

Author: claude

crates/ruvector-mincut/src/jtree/mod.rs

@@ -54,7 +54,7 @@
 //! let mut jtree = JTreeHierarchy::build(graph, config).unwrap();
 //!
 //! // Query approximate min-cut (Tier 1)
-//! let approx = jtree.approximate_min_cut();
+//! let approx = jtree.approximate_min_cut().unwrap();
 //! println!("Approximate min-cut: {} (factor: {})", approx.value, approx.approximation_factor);
 //!
 //! // Handle dynamic updates

crates/ruvector-mincut/src/optimization/benchmark.rs

@@ -165,9 +165,19 @@ impl BenchmarkSuite {
         }
 
         // Estimate combined speedup (conservative: product of square roots)
+        // Skip results with zero or negative speedup to avoid NaN
         let mut combined = 1.0;
+        let mut count = 0;
         for result in &self.results {
-            combined *= result.summary.avg_speedup.sqrt();
+            let speedup = result.summary.avg_speedup;
+            if speedup > 0.0 && speedup.is_finite() {
+                combined *= speedup.sqrt();
+                count += 1;
+            }
+        }
+
+        if count == 0 {
+            return 1.0;
         }
 
         combined
@@ -703,7 +713,10 @@ mod tests {
         suite.run_all();
         let combined = suite.combined_speedup();
 
-        assert!(combined > 0.0);
+        // For very small inputs, overhead may exceed benefit
+        // Just verify we get a valid positive result
+        assert!(combined > 0.0 && combined.is_finite(),
+            "Combined speedup {} should be positive and finite", combined);
     }
 
     #[test]

crates/ruvector-mincut/src/optimization/parallel.rs

@@ -201,7 +201,7 @@ impl ParallelLevelUpdater {
         Self {
             scheduler: Arc::new(WorkStealingScheduler::with_config(config.clone())),
             config,
-            global_min: AtomicU64::new(u64::MAX),
+            global_min: AtomicU64::new(f64::INFINITY.to_bits()),
             best_level: AtomicUsize::new(usize::MAX),
         }
     }
@@ -249,7 +249,7 @@ impl ParallelLevelUpdater {
 
     /// Reset global minimum
     pub fn reset_min(&self) {
-        self.global_min.store(u64::MAX, Ordering::Release);
+        self.global_min.store(f64::INFINITY.to_bits(), Ordering::Release);
         self.best_level.store(usize::MAX, Ordering::Release);
     }