Load edge visibility accessors as typed arrays to cut JS heap use

[markschlosseratbentley]

Description

GltfLoader.loadEdgeVisibility loaded edge-visibility accessors (visibility, silhouetteNormals, lineStrings[].indices) via loadAccessor, converting binary data into plain JS number arrays retained for the tileset's lifetime. On edge-heavy tilesets this inflated JS heap ~12x (39.5 MB -> ~460 MB on a 306-tile test dataset).

• Load all three accessors with loadAccessorTypedArray instead
• EdgeVisibilityPipelineStage now reads silhouetteNormals as a flat packed Int8Array (was Cartesian3[])
• Specs updated for the new shape; added a value-level decode test for silhouette face normals

Measured heap delta after fix: ~460 MB -> 45 MB (Tafe_Sydney), 31.5 MB -> 6.6 MB (04_Plant). FPS/GPU memory unchanged.

Note: primitive.edgeVisibility.silhouetteNormals changes shape from Cartesian3[] to flat Int8Array (internal/undocumented API).

Explanation

Plain JS number arrays store elements as V8 tagged values or heap-allocated doubles rather than raw bytes, so 1-byte binary values inflate several-fold; a heap allocation profile attributed ~340 MB of the regression to loadAccessorValues. Typed arrays store the data as a compact binary buffer, 1 byte per element.

Issue number and link

#13640

Testing plan

• All EdgeVisibility unit specs pass (gulp test --webglStub), including new typed-array assertions and a value-level silhouette-normal decode test.
• Benchmarked with a Playwright harness (fresh browser per run, forced GC, empty-scene baseline delta): on a 306-tile edge-heavy tileset, JS heap delta dropped ~460 MB -> ~45 MB; FPS, GPU bytes, and triangle counts unchanged. Control runs on unpatched builds reproduced the original numbers.
• Manually verified silhouette edge rendering in Sandcastle.

Author checklist

• I have submitted a Contributor License Agreement
• I have added my name to CONTRIBUTORS.md
• I have updated CHANGES.md with a short summary of my change
• I have added or updated unit tests to ensure consistent code coverage
• I have updated the inline documentation, and included code examples where relevant
• I have performed a self-review of my code

AI acknowledgment

• I used AI to generate content in this PR
• If yes, I have reviewed the AI-generated content before submitting

If yes, I used the following Tools(s) and/or Service(s):

GitHub Copilot

If yes, I used the following Model(s):

Claude Fable 5

[github-actions]

Thank you for the pull request, @markschlosseratbentley!

✅ We can confirm we have a CLA on file for you.