perf(modeling): Expensive string key generation for Map lookups in scissionGeom3.js

[jbroll]

Summary

scissionGeom3.js converts vec3 arrays to strings for use as Map keys, which is expensive when done repeatedly.

Usage Frequency: LOW

Only used by scission() operation.

Typical model usage: Rare - maybe 1% of models

Problem

File: packages/modeling/src/operations/booleans/scissionGeom3.js
Lines: 9-22, 35

const insertMapping = (map, point, index) => {
  const key = \`\${point}\`  // Calls toString on array, creates string
  const mapping = map.get(key)
  if (mapping === undefined) {
    map.set(key, [index])
  } else {
    mapping.push(index)
  }
}

const findMapping = (map, point) => {
  const key = \`\${point}\`  // String creation again
  return map.get(key)
}

Called in loops:

polygons.forEach((polygon, index) => {
  polygon.vertices.forEach((point) => {
    insertMapping(indexesPerPoint, vec3.snap(temp, point, eps), index)
  })
})

For a geometry with 10,000 vertices, this creates 10,000+ strings just for lookups.

Suggested Fix

Consider alternative approaches:

1. Use a spatial hash with integer keys
2. Cache the string representation on first computation
3. Use a custom Map implementation that handles vec3 keys directly

// Option 1: Spatial hash with integer key
const spatialKey = (point, eps) => {
  const scale = 1 / eps
  return \`\${Math.round(point[0] * scale)},\${Math.round(point[1] * scale)},\${Math.round(point[2] * scale)}\`
}

// This still creates strings but they're shorter and computed once per lookup

Impact

Low overall (rare operation), but contributes to overhead when scission is used.