Implement Spatial Partitioning for Large Graphs

[lowhung]

Description

The current force-directed layout uses O(n²) all-pairs repulsion calculation. For graphs with 500+ nodes, this will become a performance bottleneck.

Current State

// src/ui/layout.rs:60-82
for i in 0..nodes.len() {
    for j in (i + 1)..nodes.len() {
        // O(n²) comparison
    }
}

Proposed Solution

Implement Barnes-Hut algorithm using a quadtree:

1. Build quadtree of node positions each frame
2. For distant node clusters, approximate as single point mass
3. Only compute exact forces for nearby nodes
4. Reduces complexity to O(n log n)

Acceptance Criteria

• Implement quadtree data structure
• Implement Barnes-Hut approximation (theta parameter ~0.5)
• Add benchmark comparing O(n²) vs O(n log n) for various graph sizes
• Fallback to simple algorithm for small graphs (<100 nodes)
• No visual regression for typical graph sizes

Files Affected

• src/ui/layout.rs
• New: src/ui/layout/quadtree.rs (optional)