Replace O(n²) lookups with Map-based atoms in schema state layer

[kmcginnes]

Description

Replace O(n²) .find() loops with O(1) Map-based lookups across the schema state layer. At 70k vertex types, mergeConfiguration goes from 18.8 seconds to 162 milliseconds (116x faster).

• Add vertexPreferencesAtom and edgePreferencesAtom with .get(type) accessor that returns resolved preferences via O(1) Map lookup with default fallback
• Simplify vertexPreferenceByTypeAtom / edgePreferenceByTypeAtom to delegate to the new Map atoms
• Simplify useAllVertexPreferences / useAllEdgePreferences by removing intermediate useStoredGraphPreferences wrapper
• Pre-index vertex, edge, and preference arrays into Maps in mergeConfiguration, converting O(n²) .find() calls to O(1) .get() lookups
• Convert mergeAttributes to use Map-based lookups
• Replace lodash.cloneDeep with destructuring spread in patchToRemoveDisplayLabel, avoiding 70k deep clones
• Replace lodash.uniq() with native Set for deduplication
• Fix circular import in storageAtoms.ts

Benchmarks (mergeConfiguration, 3 attributes per type)

Types	Old (mean)	New (mean)	Speedup
20	0.048ms	0.029ms	1.7x
1,000	5.98ms	1.72ms	3.5x
10,000	506ms	20.9ms	24x
70,000	18,770ms	162ms	116x

Validation

• All existing tests pass with no changes to assertions — behavior is preserved
• New tests verify patchToRemoveDisplayLabel does not mutate its input
• New tests verify the Map-based preference atoms return stored preferences for known types and defaults for unknown types
• Benchmarked with a synthetic 70k-type schema

Related Issues

• Closes Eliminate O(n²) operations in the schema state layer #1678
• Part of Support graphs with very large numbers of node types (70k+) #1677

Check List

• I confirm that my contribution is made under the terms of the Apache 2.0 license.
• I have verified pnpm checks passes with no errors.
• I have verified pnpm test passes with no failures.
• I have covered new added functionality with unit tests if necessary.
• I have updated documentation if necessary.