Unify CodeGraphy on the WebGPU renderer

[joesobo]

Goal

Make CodeGraphy's custom WebGPU and WebAssembly graph stack the single graph renderer while keeping product behavior and renderer control in the VS Code extension.

Package boundary

• Added the focused, publishable @codegraphy-dev/graph-renderer package.
• The package owns renderer-facing contracts, WebGPU drawing and geometry, the typed-array physics engine, and the AssemblyScript/WebAssembly physics kernel.
• The extension still owns UI, settings, persistence, graph state, interaction routing, camera behavior, renderer lifecycle, fixed-timestep scheduling, and the FPS / ms toggle.
• The renderer package has no extension, React, or runtime dependency. Its root export, declarations, generated WASM, README, license, and package metadata stay owned by the package.

Physically meaningful node size

• Collision correction uses the opposite node's squared scaled radius, matching d3-force's collision weighting: small nodes move around large hubs instead of splitting every overlap equally.
• Default charge multipliers scale with semantic node size relative to size 16 and clamp to 0.5–4. Explicit plugin-provided charge multipliers remain authoritative.
• WebGPU node instances are packed in ascending drawn area so large nodes paint last without changing physics/layout indexes.
• The size order is cached and rebuilt only when drawn sizes change.
• Hovered nodes are submitted after every normal node and remain visually topmost.
• Spatial picking uses the same size-first stacking rule when hit areas overlap.

Draw-order performance

The size-order indirection is effectively free on steady frames and only sorts after a size change. Local measurements found approximately 0.0011 ms / 0.0158 ms / 0.265 ms of steady packing overhead for 10k / 100k / 1M nodes, with one-time size-change sorts of 1.06 ms / 10.91 ms / 115.24 ms. No noticeable renderer slowdown was found at expected graph sizes.

Cleanup

• Removed @codegraphy-dev/graph-benchmark, its commands, scenarios, generated reports, recordings, and reference media.
• Removed stage-attribution, interaction-recording, and benchmark-only runtime instrumentation.
• Kept the compact FPS · ms display with a bounded rolling average.
• Removed d3-force and @types/d3-force from the workspace.
• Removed stale extraction paths, duplicated WASM watch builds, obsolete renderer wording, and old renderer screenshots. All renderer-owned TypeScript, AssemblyScript, and generated WASM now live under packages/graph-renderer/src.
• Cleared every repository CRAP advisory, including the pre-existing backlog, by splitting responsibilities without changing behavior. The organization gate is clean against its refreshed baseline.
• Kept every production file changed by this PR below 50 Stryker mutation sites; renderer physics and WebGPU modules are grouped by stable behavior ownership.

Physics and renderer regression coverage

The renderer package owns deterministic tests for force primitives, radius-weighted collision movement, mixed-force trajectory hashes, TypeScript/WASM integration, 500-node graph behavior, charge-size mapping, cached instance ordering, hovered-node submission order, WASM loading/retry, device loss and cleanup, geometry, and renderer behavior. Extension tests cover plugin charge precedence and size-aware overlapping-node picking.

Validation

• Monorepo build, typecheck, and lint: green locally
• Extension unit tests: 5,756 / 5,756 across 981 files
• Core unit tests: 1,158 / 1,158 across 251 files
• Graph renderer tests: 173 / 173 across 30 files
• Renderer coverage: 97.32% statements, 89.16% branches, 98.93% functions, 98.81% lines
• Repository and renderer CRAP gates: every function at or below 8; all 65 pre-existing advisories cleared
• Whole-PR mutation-site inventory remained below the 50-site threshold; this cleanup only moved or further split renderer modules.
• Boundary/reachability gates: 2,006 files, 0 layer violations, dead surfaces, or dead ends
• Scoped physics mutation diagnostic: 189 / 196 killed (96.43%), with seven equivalent/cosmetic boundary survivors
• Scoped renderer mutation diagnostic: 60.29% overall; used to guide focused splits, with broader GPU/cache/frame-queue concerns documented for follow-up rather than churn
• The cleanup/hardening phase did not edit the human-owned acceptance Gherkin; the PR includes the FPS counter feature spec added with that product behavior.

[changeset-bot]

🦋 Changeset detected

Latest commit: 6185a96

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 12 packages

Name	Type
@codegraphy-dev/extension	Minor
@codegraphy-dev/graph-renderer	Minor
@codegraphy-dev/plugin-api	Major
@codegraphy-dev/core	Major
@codegraphy-dev/plugin-unity	Patch
@codegraphy-dev/plugin-godot	Patch
@codegraphy-dev/plugin-markdown	Patch
@codegraphy-dev/plugin-particles	Patch
@codegraphy-dev/plugin-svelte	Patch
@codegraphy-dev/plugin-typescript	Patch
@codegraphy-dev/plugin-vue	Patch
@codegraphy-dev/mcp	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR